2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
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
20 #include "xfs_types.h"
24 #include "xfs_trans.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_dir_sf.h"
36 #include "xfs_dir2_sf.h"
37 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_alloc.h"
41 #include "xfs_btree.h"
42 #include "xfs_inode_item.h"
45 #include "xfs_attr_leaf.h"
46 #include "xfs_error.h"
47 #include "xfs_quota.h"
48 #include "xfs_trans_space.h"
55 * Provide the external interfaces to manage attribute lists.
58 #define ATTR_SYSCOUNT 2
59 STATIC struct attrnames posix_acl_access;
60 STATIC struct attrnames posix_acl_default;
61 STATIC struct attrnames *attr_system_names[ATTR_SYSCOUNT];
63 /*========================================================================
64 * Function prototypes for the kernel.
65 *========================================================================*/
68 * Internal routines when attribute list fits inside the inode.
70 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
73 * Internal routines when attribute list is one block.
75 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
76 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
77 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
78 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
81 * Internal routines when attribute list is more than one block.
83 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
84 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
85 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
86 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
87 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
88 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
91 * Routines to manipulate out-of-line attribute values.
93 STATIC int xfs_attr_rmtval_get(xfs_da_args_t *args);
94 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
95 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
97 #define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
99 #if defined(XFS_ATTR_TRACE)
100 ktrace_t *xfs_attr_trace_buf;
104 /*========================================================================
105 * Overall external interface routines.
106 *========================================================================*/
109 xfs_attr_fetch(xfs_inode_t *ip, const char *name, int namelen,
110 char *value, int *valuelenp, int flags, struct cred *cred)
115 if ((XFS_IFORK_Q(ip) == 0) ||
116 (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
117 ip->i_d.di_anextents == 0))
121 * Fill in the arg structure for this request.
123 memset((char *)&args, 0, sizeof(args));
125 args.namelen = namelen;
127 args.valuelen = *valuelenp;
129 args.hashval = xfs_da_hashname(args.name, args.namelen);
131 args.whichfork = XFS_ATTR_FORK;
134 * Decide on what work routines to call based on the inode size.
136 if (XFS_IFORK_Q(ip) == 0 ||
137 (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
138 ip->i_d.di_anextents == 0)) {
139 error = XFS_ERROR(ENOATTR);
140 } else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
141 error = xfs_attr_shortform_getvalue(&args);
142 } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
143 error = xfs_attr_leaf_get(&args);
145 error = xfs_attr_node_get(&args);
149 * Return the number of bytes in the value to the caller.
151 *valuelenp = args.valuelen;
159 xfs_attr_get(bhv_desc_t *bdp, const char *name, char *value, int *valuelenp,
160 int flags, struct cred *cred)
162 xfs_inode_t *ip = XFS_BHVTOI(bdp);
165 XFS_STATS_INC(xs_attr_get);
169 namelen = strlen(name);
170 if (namelen >= MAXNAMELEN)
171 return(EFAULT); /* match IRIX behaviour */
173 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
176 xfs_ilock(ip, XFS_ILOCK_SHARED);
177 error = xfs_attr_fetch(ip, name, namelen, value, valuelenp, flags, cred);
178 xfs_iunlock(ip, XFS_ILOCK_SHARED);
183 xfs_attr_set_int(xfs_inode_t *dp, const char *name, int namelen,
184 char *value, int valuelen, int flags)
187 xfs_fsblock_t firstblock;
188 xfs_bmap_free_t flist;
189 int error, err2, committed;
192 xfs_mount_t *mp = dp->i_mount;
193 int rsvd = (flags & ATTR_ROOT) != 0;
196 * Attach the dquots to the inode.
198 if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
202 * Determine space new attribute will use, and if it would be
203 * "local" or "remote" (note: local != inline).
205 size = xfs_attr_leaf_newentsize(namelen, valuelen,
206 mp->m_sb.sb_blocksize, &local);
209 * If the inode doesn't have an attribute fork, add one.
210 * (inode must not be locked when we call this routine)
212 if (XFS_IFORK_Q(dp) == 0) {
213 if ((error = xfs_bmap_add_attrfork(dp, size, rsvd)))
218 * Fill in the arg structure for this request.
220 memset((char *)&args, 0, sizeof(args));
222 args.namelen = namelen;
224 args.valuelen = valuelen;
226 args.hashval = xfs_da_hashname(args.name, args.namelen);
228 args.firstblock = &firstblock;
230 args.whichfork = XFS_ATTR_FORK;
234 nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
236 if (size > (mp->m_sb.sb_blocksize >> 1)) {
237 /* Double split possible */
241 uint dblocks = XFS_B_TO_FSB(mp, valuelen);
242 /* Out of line attribute, cannot double split, but make
243 * room for the attribute value itself.
246 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
249 /* Size is now blocks for attribute data */
253 * Start our first transaction of the day.
255 * All future transactions during this code must be "chained" off
256 * this one via the trans_dup() call. All transactions will contain
257 * the inode, and the inode will always be marked with trans_ihold().
258 * Since the inode will be locked in all transactions, we must log
259 * the inode in every transaction to let it float upward through
262 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
265 * Root fork attributes can use reserved data blocks for this
266 * operation if necessary
270 args.trans->t_flags |= XFS_TRANS_RESERVE;
272 if ((error = xfs_trans_reserve(args.trans, (uint) nblks,
273 XFS_ATTRSET_LOG_RES(mp, nblks),
274 0, XFS_TRANS_PERM_LOG_RES,
275 XFS_ATTRSET_LOG_COUNT))) {
276 xfs_trans_cancel(args.trans, 0);
279 xfs_ilock(dp, XFS_ILOCK_EXCL);
281 error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, args.trans, dp, nblks, 0,
282 rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
283 XFS_QMOPT_RES_REGBLKS);
285 xfs_iunlock(dp, XFS_ILOCK_EXCL);
286 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
290 xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
291 xfs_trans_ihold(args.trans, dp);
294 * If the attribute list is non-existant or a shortform list,
295 * upgrade it to a single-leaf-block attribute list.
297 if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
298 ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
299 (dp->i_d.di_anextents == 0))) {
302 * Build initial attribute list (if required).
304 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
305 xfs_attr_shortform_create(&args);
308 * Try to add the attr to the attribute list in
311 error = xfs_attr_shortform_addname(&args);
312 if (error != ENOSPC) {
314 * Commit the shortform mods, and we're done.
315 * NOTE: this is also the error path (EEXIST, etc).
317 ASSERT(args.trans != NULL);
320 * If this is a synchronous mount, make sure that
321 * the transaction goes to disk before returning
324 if (mp->m_flags & XFS_MOUNT_WSYNC) {
325 xfs_trans_set_sync(args.trans);
327 err2 = xfs_trans_commit(args.trans,
328 XFS_TRANS_RELEASE_LOG_RES,
330 xfs_iunlock(dp, XFS_ILOCK_EXCL);
333 * Hit the inode change time.
335 if (!error && (flags & ATTR_KERNOTIME) == 0) {
336 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
338 return(error == 0 ? err2 : error);
342 * It won't fit in the shortform, transform to a leaf block.
343 * GROT: another possible req'mt for a double-split btree op.
345 XFS_BMAP_INIT(args.flist, args.firstblock);
346 error = xfs_attr_shortform_to_leaf(&args);
348 error = xfs_bmap_finish(&args.trans, args.flist,
349 *args.firstblock, &committed);
354 xfs_bmap_cancel(&flist);
359 * bmap_finish() may have committed the last trans and started
360 * a new one. We need the inode to be in all transactions.
363 xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
364 xfs_trans_ihold(args.trans, dp);
368 * Commit the leaf transformation. We'll need another (linked)
369 * transaction to add the new attribute to the leaf.
371 if ((error = xfs_attr_rolltrans(&args.trans, dp)))
376 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
377 error = xfs_attr_leaf_addname(&args);
379 error = xfs_attr_node_addname(&args);
386 * If this is a synchronous mount, make sure that the
387 * transaction goes to disk before returning to the user.
389 if (mp->m_flags & XFS_MOUNT_WSYNC) {
390 xfs_trans_set_sync(args.trans);
394 * Commit the last in the sequence of transactions.
396 xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
397 error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
399 xfs_iunlock(dp, XFS_ILOCK_EXCL);
402 * Hit the inode change time.
404 if (!error && (flags & ATTR_KERNOTIME) == 0) {
405 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
412 xfs_trans_cancel(args.trans,
413 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
414 xfs_iunlock(dp, XFS_ILOCK_EXCL);
419 xfs_attr_set(bhv_desc_t *bdp, const char *name, char *value, int valuelen, int flags,
425 namelen = strlen(name);
426 if (namelen >= MAXNAMELEN)
427 return EFAULT; /* match IRIX behaviour */
429 XFS_STATS_INC(xs_attr_set);
431 dp = XFS_BHVTOI(bdp);
432 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
435 return xfs_attr_set_int(dp, name, namelen, value, valuelen, flags);
439 * Generic handler routine to remove a name from an attribute list.
440 * Transitions attribute list from Btree to shortform as necessary.
443 xfs_attr_remove_int(xfs_inode_t *dp, const char *name, int namelen, int flags)
446 xfs_fsblock_t firstblock;
447 xfs_bmap_free_t flist;
449 xfs_mount_t *mp = dp->i_mount;
452 * Fill in the arg structure for this request.
454 memset((char *)&args, 0, sizeof(args));
456 args.namelen = namelen;
458 args.hashval = xfs_da_hashname(args.name, args.namelen);
460 args.firstblock = &firstblock;
463 args.whichfork = XFS_ATTR_FORK;
466 * Attach the dquots to the inode.
468 if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
472 * Start our first transaction of the day.
474 * All future transactions during this code must be "chained" off
475 * this one via the trans_dup() call. All transactions will contain
476 * the inode, and the inode will always be marked with trans_ihold().
477 * Since the inode will be locked in all transactions, we must log
478 * the inode in every transaction to let it float upward through
481 args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
484 * Root fork attributes can use reserved data blocks for this
485 * operation if necessary
488 if (flags & ATTR_ROOT)
489 args.trans->t_flags |= XFS_TRANS_RESERVE;
491 if ((error = xfs_trans_reserve(args.trans,
492 XFS_ATTRRM_SPACE_RES(mp),
493 XFS_ATTRRM_LOG_RES(mp),
494 0, XFS_TRANS_PERM_LOG_RES,
495 XFS_ATTRRM_LOG_COUNT))) {
496 xfs_trans_cancel(args.trans, 0);
500 xfs_ilock(dp, XFS_ILOCK_EXCL);
502 * No need to make quota reservations here. We expect to release some
503 * blocks not allocate in the common case.
505 xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
506 xfs_trans_ihold(args.trans, dp);
509 * Decide on what work routines to call based on the inode size.
511 if (XFS_IFORK_Q(dp) == 0 ||
512 (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
513 dp->i_d.di_anextents == 0)) {
514 error = XFS_ERROR(ENOATTR);
517 if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
518 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
519 error = xfs_attr_shortform_remove(&args);
523 } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
524 error = xfs_attr_leaf_removename(&args);
526 error = xfs_attr_node_removename(&args);
533 * If this is a synchronous mount, make sure that the
534 * transaction goes to disk before returning to the user.
536 if (mp->m_flags & XFS_MOUNT_WSYNC) {
537 xfs_trans_set_sync(args.trans);
541 * Commit the last in the sequence of transactions.
543 xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
544 error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
546 xfs_iunlock(dp, XFS_ILOCK_EXCL);
549 * Hit the inode change time.
551 if (!error && (flags & ATTR_KERNOTIME) == 0) {
552 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
559 xfs_trans_cancel(args.trans,
560 XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
561 xfs_iunlock(dp, XFS_ILOCK_EXCL);
566 xfs_attr_remove(bhv_desc_t *bdp, const char *name, int flags, struct cred *cred)
571 namelen = strlen(name);
572 if (namelen >= MAXNAMELEN)
573 return EFAULT; /* match IRIX behaviour */
575 XFS_STATS_INC(xs_attr_remove);
577 dp = XFS_BHVTOI(bdp);
578 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
581 xfs_ilock(dp, XFS_ILOCK_SHARED);
582 if (XFS_IFORK_Q(dp) == 0 ||
583 (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
584 dp->i_d.di_anextents == 0)) {
585 xfs_iunlock(dp, XFS_ILOCK_SHARED);
586 return(XFS_ERROR(ENOATTR));
588 xfs_iunlock(dp, XFS_ILOCK_SHARED);
590 return xfs_attr_remove_int(dp, name, namelen, flags);
594 * Generate a list of extended attribute names and optionally
595 * also value lengths. Positive return value follows the XFS
596 * convention of being an error, zero or negative return code
597 * is the length of the buffer returned (negated), indicating
601 xfs_attr_list(bhv_desc_t *bdp, char *buffer, int bufsize, int flags,
602 attrlist_cursor_kern_t *cursor, struct cred *cred)
604 xfs_attr_list_context_t context;
608 XFS_STATS_INC(xs_attr_list);
611 * Validate the cursor.
613 if (cursor->pad1 || cursor->pad2)
614 return(XFS_ERROR(EINVAL));
615 if ((cursor->initted == 0) &&
616 (cursor->hashval || cursor->blkno || cursor->offset))
617 return(XFS_ERROR(EINVAL));
620 * Check for a properly aligned buffer.
622 if (((long)buffer) & (sizeof(int)-1))
623 return(XFS_ERROR(EFAULT));
624 if (flags & ATTR_KERNOVAL)
628 * Initialize the output buffer.
630 context.dp = dp = XFS_BHVTOI(bdp);
631 context.cursor = cursor;
635 context.flags = flags;
636 if (!(flags & ATTR_KERNAMELS)) {
637 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
638 context.firstu = context.bufsize;
639 context.alist = (attrlist_t *)buffer;
640 context.alist->al_count = 0;
641 context.alist->al_more = 0;
642 context.alist->al_offset[0] = context.bufsize;
645 context.bufsize = bufsize;
646 context.firstu = context.bufsize;
647 context.alist = (attrlist_t *)buffer;
650 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
653 xfs_ilock(dp, XFS_ILOCK_SHARED);
655 * Decide on what work routines to call based on the inode size.
657 xfs_attr_trace_l_c("syscall start", &context);
658 if (XFS_IFORK_Q(dp) == 0 ||
659 (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
660 dp->i_d.di_anextents == 0)) {
662 } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
663 error = xfs_attr_shortform_list(&context);
664 } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
665 error = xfs_attr_leaf_list(&context);
667 error = xfs_attr_node_list(&context);
669 xfs_iunlock(dp, XFS_ILOCK_SHARED);
670 xfs_attr_trace_l_c("syscall end", &context);
672 if (!(context.flags & (ATTR_KERNOVAL|ATTR_KERNAMELS))) {
675 else { /* must return negated buffer size or the error */
676 if (context.count < 0)
677 error = XFS_ERROR(ERANGE);
679 error = -context.count;
686 xfs_attr_inactive(xfs_inode_t *dp)
693 ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
695 xfs_ilock(dp, XFS_ILOCK_SHARED);
696 if ((XFS_IFORK_Q(dp) == 0) ||
697 (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
698 (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
699 dp->i_d.di_anextents == 0)) {
700 xfs_iunlock(dp, XFS_ILOCK_SHARED);
703 xfs_iunlock(dp, XFS_ILOCK_SHARED);
706 * Start our first transaction of the day.
708 * All future transactions during this code must be "chained" off
709 * this one via the trans_dup() call. All transactions will contain
710 * the inode, and the inode will always be marked with trans_ihold().
711 * Since the inode will be locked in all transactions, we must log
712 * the inode in every transaction to let it float upward through
715 trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
716 if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
717 XFS_TRANS_PERM_LOG_RES,
718 XFS_ATTRINVAL_LOG_COUNT))) {
719 xfs_trans_cancel(trans, 0);
722 xfs_ilock(dp, XFS_ILOCK_EXCL);
725 * No need to make quota reservations here. We expect to release some
726 * blocks, not allocate, in the common case.
728 xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
729 xfs_trans_ihold(trans, dp);
732 * Decide on what work routines to call based on the inode size.
734 if ((XFS_IFORK_Q(dp) == 0) ||
735 (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
736 (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
737 dp->i_d.di_anextents == 0)) {
741 error = xfs_attr_root_inactive(&trans, dp);
745 * signal synchronous inactive transactions unless this
746 * is a synchronous mount filesystem in which case we
747 * know that we're here because we've been called out of
748 * xfs_inactive which means that the last reference is gone
749 * and the unlink transaction has already hit the disk so
750 * async inactive transactions are safe.
752 if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
753 (!(mp->m_flags & XFS_MOUNT_WSYNC)
758 * Commit the last in the sequence of transactions.
760 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
761 error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES,
763 xfs_iunlock(dp, XFS_ILOCK_EXCL);
768 xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
769 xfs_iunlock(dp, XFS_ILOCK_EXCL);
775 /*========================================================================
776 * External routines when attribute list is inside the inode
777 *========================================================================*/
780 * Add a name to the shortform attribute list structure
781 * This is the external routine.
784 xfs_attr_shortform_addname(xfs_da_args_t *args)
786 int newsize, forkoff, retval;
788 retval = xfs_attr_shortform_lookup(args);
789 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
791 } else if (retval == EEXIST) {
792 if (args->flags & ATTR_CREATE)
794 retval = xfs_attr_shortform_remove(args);
798 if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
799 args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
800 return(XFS_ERROR(ENOSPC));
802 newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
803 newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
805 forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
807 return(XFS_ERROR(ENOSPC));
809 xfs_attr_shortform_add(args, forkoff);
814 /*========================================================================
815 * External routines when attribute list is one block
816 *========================================================================*/
819 * Add a name to the leaf attribute list structure
821 * This leaf block cannot have a "remote" value, we only call this routine
822 * if bmap_one_block() says there is only one block (ie: no remote blks).
825 xfs_attr_leaf_addname(xfs_da_args_t *args)
829 int retval, error, committed, forkoff;
832 * Read the (only) block in the attribute list in.
836 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
843 * Look up the given attribute in the leaf block. Figure out if
844 * the given flags produce an error or call for an atomic rename.
846 retval = xfs_attr_leaf_lookup_int(bp, args);
847 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
848 xfs_da_brelse(args->trans, bp);
850 } else if (retval == EEXIST) {
851 if (args->flags & ATTR_CREATE) { /* pure create op */
852 xfs_da_brelse(args->trans, bp);
855 args->rename = 1; /* an atomic rename */
856 args->blkno2 = args->blkno; /* set 2nd entry info*/
857 args->index2 = args->index;
858 args->rmtblkno2 = args->rmtblkno;
859 args->rmtblkcnt2 = args->rmtblkcnt;
863 * Add the attribute to the leaf block, transitioning to a Btree
866 retval = xfs_attr_leaf_add(bp, args);
868 if (retval == ENOSPC) {
870 * Promote the attribute list to the Btree format, then
871 * Commit that transaction so that the node_addname() call
872 * can manage its own transactions.
874 XFS_BMAP_INIT(args->flist, args->firstblock);
875 error = xfs_attr_leaf_to_node(args);
877 error = xfs_bmap_finish(&args->trans, args->flist,
878 *args->firstblock, &committed);
883 xfs_bmap_cancel(args->flist);
888 * bmap_finish() may have committed the last trans and started
889 * a new one. We need the inode to be in all transactions.
892 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
893 xfs_trans_ihold(args->trans, dp);
897 * Commit the current trans (including the inode) and start
900 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
904 * Fob the whole rest of the problem off on the Btree code.
906 error = xfs_attr_node_addname(args);
911 * Commit the transaction that added the attr name so that
912 * later routines can manage their own transactions.
914 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
918 * If there was an out-of-line value, allocate the blocks we
919 * identified for its storage and copy the value. This is done
920 * after we create the attribute so that we don't overflow the
921 * maximum size of a transaction and/or hit a deadlock.
923 if (args->rmtblkno > 0) {
924 error = xfs_attr_rmtval_set(args);
930 * If this is an atomic rename operation, we must "flip" the
931 * incomplete flags on the "new" and "old" attribute/value pairs
932 * so that one disappears and one appears atomically. Then we
933 * must remove the "old" attribute/value pair.
937 * In a separate transaction, set the incomplete flag on the
938 * "old" attr and clear the incomplete flag on the "new" attr.
940 error = xfs_attr_leaf_flipflags(args);
945 * Dismantle the "old" attribute/value pair by removing
946 * a "remote" value (if it exists).
948 args->index = args->index2;
949 args->blkno = args->blkno2;
950 args->rmtblkno = args->rmtblkno2;
951 args->rmtblkcnt = args->rmtblkcnt2;
952 if (args->rmtblkno) {
953 error = xfs_attr_rmtval_remove(args);
959 * Read in the block containing the "old" attr, then
960 * remove the "old" attr from that block (neat, huh!)
962 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
967 (void)xfs_attr_leaf_remove(bp, args);
970 * If the result is small enough, shrink it all into the inode.
972 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
973 XFS_BMAP_INIT(args->flist, args->firstblock);
974 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
975 /* bp is gone due to xfs_da_shrink_inode */
977 error = xfs_bmap_finish(&args->trans,
985 xfs_bmap_cancel(args->flist);
990 * bmap_finish() may have committed the last trans
991 * and started a new one. We need the inode to be
992 * in all transactions.
995 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
996 xfs_trans_ihold(args->trans, dp);
1002 * Commit the remove and start the next trans in series.
1004 error = xfs_attr_rolltrans(&args->trans, dp);
1006 } else if (args->rmtblkno > 0) {
1008 * Added a "remote" value, just clear the incomplete flag.
1010 error = xfs_attr_leaf_clearflag(args);
1016 * Remove a name from the leaf attribute list structure
1018 * This leaf block cannot have a "remote" value, we only call this routine
1019 * if bmap_one_block() says there is only one block (ie: no remote blks).
1022 xfs_attr_leaf_removename(xfs_da_args_t *args)
1026 int error, committed, forkoff;
1029 * Remove the attribute.
1033 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1040 error = xfs_attr_leaf_lookup_int(bp, args);
1041 if (error == ENOATTR) {
1042 xfs_da_brelse(args->trans, bp);
1046 (void)xfs_attr_leaf_remove(bp, args);
1049 * If the result is small enough, shrink it all into the inode.
1051 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1052 XFS_BMAP_INIT(args->flist, args->firstblock);
1053 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1054 /* bp is gone due to xfs_da_shrink_inode */
1056 error = xfs_bmap_finish(&args->trans, args->flist,
1057 *args->firstblock, &committed);
1062 xfs_bmap_cancel(args->flist);
1067 * bmap_finish() may have committed the last trans and started
1068 * a new one. We need the inode to be in all transactions.
1071 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1072 xfs_trans_ihold(args->trans, dp);
1075 xfs_da_buf_done(bp);
1080 * Look up a name in a leaf attribute list structure.
1082 * This leaf block cannot have a "remote" value, we only call this routine
1083 * if bmap_one_block() says there is only one block (ie: no remote blks).
1086 xfs_attr_leaf_get(xfs_da_args_t *args)
1092 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1098 error = xfs_attr_leaf_lookup_int(bp, args);
1099 if (error != EEXIST) {
1100 xfs_da_brelse(args->trans, bp);
1103 error = xfs_attr_leaf_getvalue(bp, args);
1104 xfs_da_brelse(args->trans, bp);
1105 if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1106 error = xfs_attr_rmtval_get(args);
1112 * Copy out attribute entries for attr_list(), for leaf attribute lists.
1115 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1117 xfs_attr_leafblock_t *leaf;
1121 context->cursor->blkno = 0;
1122 error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1127 if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1128 != XFS_ATTR_LEAF_MAGIC)) {
1129 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1130 context->dp->i_mount, leaf);
1131 xfs_da_brelse(NULL, bp);
1132 return(XFS_ERROR(EFSCORRUPTED));
1135 (void)xfs_attr_leaf_list_int(bp, context);
1136 xfs_da_brelse(NULL, bp);
1141 /*========================================================================
1142 * External routines when attribute list size > XFS_LBSIZE(mp).
1143 *========================================================================*/
1146 * Add a name to a Btree-format attribute list.
1148 * This will involve walking down the Btree, and may involve splitting
1149 * leaf nodes and even splitting intermediate nodes up to and including
1150 * the root node (a special case of an intermediate node).
1152 * "Remote" attribute values confuse the issue and atomic rename operations
1153 * add a whole extra layer of confusion on top of that.
1156 xfs_attr_node_addname(xfs_da_args_t *args)
1158 xfs_da_state_t *state;
1159 xfs_da_state_blk_t *blk;
1162 int committed, retval, error;
1165 * Fill in bucket of arguments/results/context to carry around.
1170 state = xfs_da_state_alloc();
1173 state->blocksize = state->mp->m_sb.sb_blocksize;
1174 state->node_ents = state->mp->m_attr_node_ents;
1177 * Search to see if name already exists, and get back a pointer
1178 * to where it should go.
1180 error = xfs_da_node_lookup_int(state, &retval);
1183 blk = &state->path.blk[ state->path.active-1 ];
1184 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1185 if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1187 } else if (retval == EEXIST) {
1188 if (args->flags & ATTR_CREATE)
1190 args->rename = 1; /* atomic rename op */
1191 args->blkno2 = args->blkno; /* set 2nd entry info*/
1192 args->index2 = args->index;
1193 args->rmtblkno2 = args->rmtblkno;
1194 args->rmtblkcnt2 = args->rmtblkcnt;
1196 args->rmtblkcnt = 0;
1199 retval = xfs_attr_leaf_add(blk->bp, state->args);
1200 if (retval == ENOSPC) {
1201 if (state->path.active == 1) {
1203 * Its really a single leaf node, but it had
1204 * out-of-line values so it looked like it *might*
1205 * have been a b-tree.
1207 xfs_da_state_free(state);
1208 XFS_BMAP_INIT(args->flist, args->firstblock);
1209 error = xfs_attr_leaf_to_node(args);
1211 error = xfs_bmap_finish(&args->trans,
1219 xfs_bmap_cancel(args->flist);
1224 * bmap_finish() may have committed the last trans
1225 * and started a new one. We need the inode to be
1226 * in all transactions.
1229 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1230 xfs_trans_ihold(args->trans, dp);
1234 * Commit the node conversion and start the next
1235 * trans in the chain.
1237 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1244 * Split as many Btree elements as required.
1245 * This code tracks the new and old attr's location
1246 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1247 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1249 XFS_BMAP_INIT(args->flist, args->firstblock);
1250 error = xfs_da_split(state);
1252 error = xfs_bmap_finish(&args->trans, args->flist,
1253 *args->firstblock, &committed);
1258 xfs_bmap_cancel(args->flist);
1263 * bmap_finish() may have committed the last trans and started
1264 * a new one. We need the inode to be in all transactions.
1267 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1268 xfs_trans_ihold(args->trans, dp);
1272 * Addition succeeded, update Btree hashvals.
1274 xfs_da_fixhashpath(state, &state->path);
1278 * Kill the state structure, we're done with it and need to
1279 * allow the buffers to come back later.
1281 xfs_da_state_free(state);
1285 * Commit the leaf addition or btree split and start the next
1286 * trans in the chain.
1288 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1292 * If there was an out-of-line value, allocate the blocks we
1293 * identified for its storage and copy the value. This is done
1294 * after we create the attribute so that we don't overflow the
1295 * maximum size of a transaction and/or hit a deadlock.
1297 if (args->rmtblkno > 0) {
1298 error = xfs_attr_rmtval_set(args);
1304 * If this is an atomic rename operation, we must "flip" the
1305 * incomplete flags on the "new" and "old" attribute/value pairs
1306 * so that one disappears and one appears atomically. Then we
1307 * must remove the "old" attribute/value pair.
1311 * In a separate transaction, set the incomplete flag on the
1312 * "old" attr and clear the incomplete flag on the "new" attr.
1314 error = xfs_attr_leaf_flipflags(args);
1319 * Dismantle the "old" attribute/value pair by removing
1320 * a "remote" value (if it exists).
1322 args->index = args->index2;
1323 args->blkno = args->blkno2;
1324 args->rmtblkno = args->rmtblkno2;
1325 args->rmtblkcnt = args->rmtblkcnt2;
1326 if (args->rmtblkno) {
1327 error = xfs_attr_rmtval_remove(args);
1333 * Re-find the "old" attribute entry after any split ops.
1334 * The INCOMPLETE flag means that we will find the "old"
1335 * attr, not the "new" one.
1337 args->flags |= XFS_ATTR_INCOMPLETE;
1338 state = xfs_da_state_alloc();
1341 state->blocksize = state->mp->m_sb.sb_blocksize;
1342 state->node_ents = state->mp->m_attr_node_ents;
1344 error = xfs_da_node_lookup_int(state, &retval);
1349 * Remove the name and update the hashvals in the tree.
1351 blk = &state->path.blk[ state->path.active-1 ];
1352 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1353 error = xfs_attr_leaf_remove(blk->bp, args);
1354 xfs_da_fixhashpath(state, &state->path);
1357 * Check to see if the tree needs to be collapsed.
1359 if (retval && (state->path.active > 1)) {
1360 XFS_BMAP_INIT(args->flist, args->firstblock);
1361 error = xfs_da_join(state);
1363 error = xfs_bmap_finish(&args->trans,
1371 xfs_bmap_cancel(args->flist);
1376 * bmap_finish() may have committed the last trans
1377 * and started a new one. We need the inode to be
1378 * in all transactions.
1381 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1382 xfs_trans_ihold(args->trans, dp);
1387 * Commit and start the next trans in the chain.
1389 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1392 } else if (args->rmtblkno > 0) {
1394 * Added a "remote" value, just clear the incomplete flag.
1396 error = xfs_attr_leaf_clearflag(args);
1404 xfs_da_state_free(state);
1411 * Remove a name from a B-tree attribute list.
1413 * This will involve walking down the Btree, and may involve joining
1414 * leaf nodes and even joining intermediate nodes up to and including
1415 * the root node (a special case of an intermediate node).
1418 xfs_attr_node_removename(xfs_da_args_t *args)
1420 xfs_da_state_t *state;
1421 xfs_da_state_blk_t *blk;
1424 int retval, error, committed, forkoff;
1427 * Tie a string around our finger to remind us where we are.
1430 state = xfs_da_state_alloc();
1432 state->mp = dp->i_mount;
1433 state->blocksize = state->mp->m_sb.sb_blocksize;
1434 state->node_ents = state->mp->m_attr_node_ents;
1437 * Search to see if name exists, and get back a pointer to it.
1439 error = xfs_da_node_lookup_int(state, &retval);
1440 if (error || (retval != EEXIST)) {
1447 * If there is an out-of-line value, de-allocate the blocks.
1448 * This is done before we remove the attribute so that we don't
1449 * overflow the maximum size of a transaction and/or hit a deadlock.
1451 blk = &state->path.blk[ state->path.active-1 ];
1452 ASSERT(blk->bp != NULL);
1453 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1454 if (args->rmtblkno > 0) {
1456 * Fill in disk block numbers in the state structure
1457 * so that we can get the buffers back after we commit
1458 * several transactions in the following calls.
1460 error = xfs_attr_fillstate(state);
1465 * Mark the attribute as INCOMPLETE, then bunmapi() the
1468 error = xfs_attr_leaf_setflag(args);
1471 error = xfs_attr_rmtval_remove(args);
1476 * Refill the state structure with buffers, the prior calls
1477 * released our buffers.
1479 error = xfs_attr_refillstate(state);
1485 * Remove the name and update the hashvals in the tree.
1487 blk = &state->path.blk[ state->path.active-1 ];
1488 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1489 retval = xfs_attr_leaf_remove(blk->bp, args);
1490 xfs_da_fixhashpath(state, &state->path);
1493 * Check to see if the tree needs to be collapsed.
1495 if (retval && (state->path.active > 1)) {
1496 XFS_BMAP_INIT(args->flist, args->firstblock);
1497 error = xfs_da_join(state);
1499 error = xfs_bmap_finish(&args->trans, args->flist,
1500 *args->firstblock, &committed);
1505 xfs_bmap_cancel(args->flist);
1510 * bmap_finish() may have committed the last trans and started
1511 * a new one. We need the inode to be in all transactions.
1514 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1515 xfs_trans_ihold(args->trans, dp);
1519 * Commit the Btree join operation and start a new trans.
1521 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1526 * If the result is small enough, push it all into the inode.
1528 if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1530 * Have to get rid of the copy of this dabuf in the state.
1532 ASSERT(state->path.active == 1);
1533 ASSERT(state->path.blk[0].bp);
1534 xfs_da_buf_done(state->path.blk[0].bp);
1535 state->path.blk[0].bp = NULL;
1537 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1541 ASSERT(INT_GET(((xfs_attr_leafblock_t *)
1542 bp->data)->hdr.info.magic, ARCH_CONVERT)
1543 == XFS_ATTR_LEAF_MAGIC);
1545 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1546 XFS_BMAP_INIT(args->flist, args->firstblock);
1547 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1548 /* bp is gone due to xfs_da_shrink_inode */
1550 error = xfs_bmap_finish(&args->trans,
1558 xfs_bmap_cancel(args->flist);
1563 * bmap_finish() may have committed the last trans
1564 * and started a new one. We need the inode to be
1565 * in all transactions.
1568 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1569 xfs_trans_ihold(args->trans, dp);
1572 xfs_da_brelse(args->trans, bp);
1577 xfs_da_state_free(state);
1582 * Fill in the disk block numbers in the state structure for the buffers
1583 * that are attached to the state structure.
1584 * This is done so that we can quickly reattach ourselves to those buffers
1585 * after some set of transaction commit's has released these buffers.
1588 xfs_attr_fillstate(xfs_da_state_t *state)
1590 xfs_da_state_path_t *path;
1591 xfs_da_state_blk_t *blk;
1595 * Roll down the "path" in the state structure, storing the on-disk
1596 * block number for those buffers in the "path".
1598 path = &state->path;
1599 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1600 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1602 blk->disk_blkno = xfs_da_blkno(blk->bp);
1603 xfs_da_buf_done(blk->bp);
1606 blk->disk_blkno = 0;
1611 * Roll down the "altpath" in the state structure, storing the on-disk
1612 * block number for those buffers in the "altpath".
1614 path = &state->altpath;
1615 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1616 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1618 blk->disk_blkno = xfs_da_blkno(blk->bp);
1619 xfs_da_buf_done(blk->bp);
1622 blk->disk_blkno = 0;
1630 * Reattach the buffers to the state structure based on the disk block
1631 * numbers stored in the state structure.
1632 * This is done after some set of transaction commit's has released those
1633 * buffers from our grip.
1636 xfs_attr_refillstate(xfs_da_state_t *state)
1638 xfs_da_state_path_t *path;
1639 xfs_da_state_blk_t *blk;
1643 * Roll down the "path" in the state structure, storing the on-disk
1644 * block number for those buffers in the "path".
1646 path = &state->path;
1647 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1648 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1649 if (blk->disk_blkno) {
1650 error = xfs_da_read_buf(state->args->trans,
1652 blk->blkno, blk->disk_blkno,
1653 &blk->bp, XFS_ATTR_FORK);
1662 * Roll down the "altpath" in the state structure, storing the on-disk
1663 * block number for those buffers in the "altpath".
1665 path = &state->altpath;
1666 ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1667 for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1668 if (blk->disk_blkno) {
1669 error = xfs_da_read_buf(state->args->trans,
1671 blk->blkno, blk->disk_blkno,
1672 &blk->bp, XFS_ATTR_FORK);
1684 * Look up a filename in a node attribute list.
1686 * This routine gets called for any attribute fork that has more than one
1687 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1688 * "remote" values taking up more blocks.
1691 xfs_attr_node_get(xfs_da_args_t *args)
1693 xfs_da_state_t *state;
1694 xfs_da_state_blk_t *blk;
1698 state = xfs_da_state_alloc();
1700 state->mp = args->dp->i_mount;
1701 state->blocksize = state->mp->m_sb.sb_blocksize;
1702 state->node_ents = state->mp->m_attr_node_ents;
1705 * Search to see if name exists, and get back a pointer to it.
1707 error = xfs_da_node_lookup_int(state, &retval);
1710 } else if (retval == EEXIST) {
1711 blk = &state->path.blk[ state->path.active-1 ];
1712 ASSERT(blk->bp != NULL);
1713 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1716 * Get the value, local or "remote"
1718 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1719 if (!retval && (args->rmtblkno > 0)
1720 && !(args->flags & ATTR_KERNOVAL)) {
1721 retval = xfs_attr_rmtval_get(args);
1726 * If not in a transaction, we have to release all the buffers.
1728 for (i = 0; i < state->path.active; i++) {
1729 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1730 state->path.blk[i].bp = NULL;
1733 xfs_da_state_free(state);
1737 STATIC int /* error */
1738 xfs_attr_node_list(xfs_attr_list_context_t *context)
1740 attrlist_cursor_kern_t *cursor;
1741 xfs_attr_leafblock_t *leaf;
1742 xfs_da_intnode_t *node;
1743 xfs_da_node_entry_t *btree;
1747 cursor = context->cursor;
1748 cursor->initted = 1;
1751 * Do all sorts of validation on the passed-in cursor structure.
1752 * If anything is amiss, ignore the cursor and look up the hashval
1753 * starting from the btree root.
1756 if (cursor->blkno > 0) {
1757 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1758 &bp, XFS_ATTR_FORK);
1759 if ((error != 0) && (error != EFSCORRUPTED))
1763 switch (INT_GET(node->hdr.info.magic, ARCH_CONVERT)) {
1764 case XFS_DA_NODE_MAGIC:
1765 xfs_attr_trace_l_cn("wrong blk", context, node);
1766 xfs_da_brelse(NULL, bp);
1769 case XFS_ATTR_LEAF_MAGIC:
1771 if (cursor->hashval >
1772 INT_GET(leaf->entries[
1773 INT_GET(leaf->hdr.count,
1774 ARCH_CONVERT)-1].hashval,
1776 xfs_attr_trace_l_cl("wrong blk",
1778 xfs_da_brelse(NULL, bp);
1780 } else if (cursor->hashval <=
1781 INT_GET(leaf->entries[0].hashval,
1783 xfs_attr_trace_l_cl("maybe wrong blk",
1785 xfs_da_brelse(NULL, bp);
1790 xfs_attr_trace_l_c("wrong blk - ??", context);
1791 xfs_da_brelse(NULL, bp);
1798 * We did not find what we expected given the cursor's contents,
1799 * so we start from the top and work down based on the hash value.
1800 * Note that start of node block is same as start of leaf block.
1805 error = xfs_da_read_buf(NULL, context->dp,
1806 cursor->blkno, -1, &bp,
1810 if (unlikely(bp == NULL)) {
1811 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1813 context->dp->i_mount);
1814 return(XFS_ERROR(EFSCORRUPTED));
1817 if (INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1818 == XFS_ATTR_LEAF_MAGIC)
1820 if (unlikely(INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1821 != XFS_DA_NODE_MAGIC)) {
1822 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1824 context->dp->i_mount,
1826 xfs_da_brelse(NULL, bp);
1827 return(XFS_ERROR(EFSCORRUPTED));
1829 btree = node->btree;
1831 i < INT_GET(node->hdr.count, ARCH_CONVERT);
1834 <= INT_GET(btree->hashval,
1836 cursor->blkno = INT_GET(btree->before, ARCH_CONVERT);
1837 xfs_attr_trace_l_cb("descending",
1842 if (i == INT_GET(node->hdr.count, ARCH_CONVERT)) {
1843 xfs_da_brelse(NULL, bp);
1846 xfs_da_brelse(NULL, bp);
1852 * Roll upward through the blocks, processing each leaf block in
1853 * order. As long as there is space in the result buffer, keep
1854 * adding the information.
1858 if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1859 != XFS_ATTR_LEAF_MAGIC)) {
1860 XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1862 context->dp->i_mount, leaf);
1863 xfs_da_brelse(NULL, bp);
1864 return(XFS_ERROR(EFSCORRUPTED));
1866 error = xfs_attr_leaf_list_int(bp, context);
1867 if (error || !leaf->hdr.info.forw)
1868 break; /* not really an error, buffer full or EOF */
1869 cursor->blkno = INT_GET(leaf->hdr.info.forw, ARCH_CONVERT);
1870 xfs_da_brelse(NULL, bp);
1871 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1872 &bp, XFS_ATTR_FORK);
1875 if (unlikely((bp == NULL))) {
1876 XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1878 context->dp->i_mount);
1879 return(XFS_ERROR(EFSCORRUPTED));
1882 xfs_da_brelse(NULL, bp);
1887 /*========================================================================
1888 * External routines for manipulating out-of-line attribute values.
1889 *========================================================================*/
1892 * Read the value associated with an attribute from the out-of-line buffer
1893 * that we stored it in.
1896 xfs_attr_rmtval_get(xfs_da_args_t *args)
1898 xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1903 int nmap, error, tmp, valuelen, blkcnt, i;
1906 ASSERT(!(args->flags & ATTR_KERNOVAL));
1908 mp = args->dp->i_mount;
1910 valuelen = args->valuelen;
1911 lblkno = args->rmtblkno;
1912 while (valuelen > 0) {
1913 nmap = ATTR_RMTVALUE_MAPSIZE;
1914 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1916 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1917 NULL, 0, map, &nmap, NULL);
1922 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1923 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1924 (map[i].br_startblock != HOLESTARTBLOCK));
1925 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1926 blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1927 error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
1928 blkcnt, XFS_BUF_LOCK, &bp);
1932 tmp = (valuelen < XFS_BUF_SIZE(bp))
1933 ? valuelen : XFS_BUF_SIZE(bp);
1934 xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
1939 lblkno += map[i].br_blockcount;
1942 ASSERT(valuelen == 0);
1947 * Write the value associated with an attribute into the out-of-line buffer
1948 * that we have defined for it.
1951 xfs_attr_rmtval_set(xfs_da_args_t *args)
1954 xfs_fileoff_t lfileoff;
1956 xfs_bmbt_irec_t map;
1961 int blkcnt, valuelen, nmap, error, tmp, committed;
1968 * Find a "hole" in the attribute address space large enough for
1969 * us to drop the new attribute's value into.
1971 blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1973 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
1978 args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
1979 args->rmtblkcnt = blkcnt;
1982 * Roll through the "value", allocating blocks on disk as required.
1984 while (blkcnt > 0) {
1986 * Allocate a single extent, up to the size of the value.
1988 XFS_BMAP_INIT(args->flist, args->firstblock);
1990 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
1992 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
1994 args->firstblock, args->total, &map, &nmap,
1997 error = xfs_bmap_finish(&args->trans, args->flist,
1998 *args->firstblock, &committed);
2003 xfs_bmap_cancel(args->flist);
2008 * bmap_finish() may have committed the last trans and started
2009 * a new one. We need the inode to be in all transactions.
2012 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2013 xfs_trans_ihold(args->trans, dp);
2017 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2018 (map.br_startblock != HOLESTARTBLOCK));
2019 lblkno += map.br_blockcount;
2020 blkcnt -= map.br_blockcount;
2023 * Start the next trans in the chain.
2025 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
2030 * Roll through the "value", copying the attribute value to the
2031 * already-allocated blocks. Blocks are written synchronously
2032 * so that we can know they are all on disk before we turn off
2033 * the INCOMPLETE flag.
2035 lblkno = args->rmtblkno;
2036 valuelen = args->valuelen;
2037 while (valuelen > 0) {
2039 * Try to remember where we decided to put the value.
2041 XFS_BMAP_INIT(args->flist, args->firstblock);
2043 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2045 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2046 args->firstblock, 0, &map, &nmap, NULL);
2051 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2052 (map.br_startblock != HOLESTARTBLOCK));
2054 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2055 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2057 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno,
2058 blkcnt, XFS_BUF_LOCK);
2060 ASSERT(!XFS_BUF_GETERROR(bp));
2062 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2064 xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2065 if (tmp < XFS_BUF_SIZE(bp))
2066 xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2067 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2073 lblkno += map.br_blockcount;
2075 ASSERT(valuelen == 0);
2080 * Remove the value associated with an attribute by deleting the
2081 * out-of-line buffer that it is stored on.
2084 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2087 xfs_bmbt_irec_t map;
2091 int valuelen, blkcnt, nmap, error, done, committed;
2093 mp = args->dp->i_mount;
2096 * Roll through the "value", invalidating the attribute value's
2099 lblkno = args->rmtblkno;
2100 valuelen = args->rmtblkcnt;
2101 while (valuelen > 0) {
2103 * Try to remember where we decided to put the value.
2105 XFS_BMAP_INIT(args->flist, args->firstblock);
2107 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2109 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2110 args->firstblock, 0, &map, &nmap,
2116 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2117 (map.br_startblock != HOLESTARTBLOCK));
2119 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2120 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2123 * If the "remote" value is in the cache, remove it.
2125 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt,
2126 XFS_INCORE_TRYLOCK);
2129 XFS_BUF_UNDELAYWRITE(bp);
2134 valuelen -= map.br_blockcount;
2136 lblkno += map.br_blockcount;
2140 * Keep de-allocating extents until the remote-value region is gone.
2142 lblkno = args->rmtblkno;
2143 blkcnt = args->rmtblkcnt;
2146 XFS_BMAP_INIT(args->flist, args->firstblock);
2147 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2148 XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2149 1, args->firstblock, args->flist, &done);
2151 error = xfs_bmap_finish(&args->trans, args->flist,
2152 *args->firstblock, &committed);
2157 xfs_bmap_cancel(args->flist);
2162 * bmap_finish() may have committed the last trans and started
2163 * a new one. We need the inode to be in all transactions.
2166 xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2167 xfs_trans_ihold(args->trans, args->dp);
2171 * Close out trans and start the next one in the chain.
2173 if ((error = xfs_attr_rolltrans(&args->trans, args->dp)))
2179 #if defined(XFS_ATTR_TRACE)
2181 * Add a trace buffer entry for an attr_list context structure.
2184 xfs_attr_trace_l_c(char *where, struct xfs_attr_list_context *context)
2186 xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_C, where,
2187 (__psunsigned_t)context->dp,
2188 (__psunsigned_t)context->cursor->hashval,
2189 (__psunsigned_t)context->cursor->blkno,
2190 (__psunsigned_t)context->cursor->offset,
2191 (__psunsigned_t)context->alist,
2192 (__psunsigned_t)context->bufsize,
2193 (__psunsigned_t)context->count,
2194 (__psunsigned_t)context->firstu,
2196 ((context->count > 0) &&
2197 !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2198 ? (ATTR_ENTRY(context->alist,
2199 context->count-1)->a_valuelen)
2201 (__psunsigned_t)context->dupcnt,
2202 (__psunsigned_t)context->flags,
2203 (__psunsigned_t)NULL,
2204 (__psunsigned_t)NULL,
2205 (__psunsigned_t)NULL);
2209 * Add a trace buffer entry for a context structure and a Btree node.
2212 xfs_attr_trace_l_cn(char *where, struct xfs_attr_list_context *context,
2213 struct xfs_da_intnode *node)
2215 xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CN, where,
2216 (__psunsigned_t)context->dp,
2217 (__psunsigned_t)context->cursor->hashval,
2218 (__psunsigned_t)context->cursor->blkno,
2219 (__psunsigned_t)context->cursor->offset,
2220 (__psunsigned_t)context->alist,
2221 (__psunsigned_t)context->bufsize,
2222 (__psunsigned_t)context->count,
2223 (__psunsigned_t)context->firstu,
2225 ((context->count > 0) &&
2226 !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2227 ? (ATTR_ENTRY(context->alist,
2228 context->count-1)->a_valuelen)
2230 (__psunsigned_t)context->dupcnt,
2231 (__psunsigned_t)context->flags,
2232 (__psunsigned_t)INT_GET(node->hdr.count, ARCH_CONVERT),
2233 (__psunsigned_t)INT_GET(node->btree[0].hashval, ARCH_CONVERT),
2234 (__psunsigned_t)INT_GET(node->btree[INT_GET(node->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2238 * Add a trace buffer entry for a context structure and a Btree element.
2241 xfs_attr_trace_l_cb(char *where, struct xfs_attr_list_context *context,
2242 struct xfs_da_node_entry *btree)
2244 xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CB, where,
2245 (__psunsigned_t)context->dp,
2246 (__psunsigned_t)context->cursor->hashval,
2247 (__psunsigned_t)context->cursor->blkno,
2248 (__psunsigned_t)context->cursor->offset,
2249 (__psunsigned_t)context->alist,
2250 (__psunsigned_t)context->bufsize,
2251 (__psunsigned_t)context->count,
2252 (__psunsigned_t)context->firstu,
2254 ((context->count > 0) &&
2255 !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2256 ? (ATTR_ENTRY(context->alist,
2257 context->count-1)->a_valuelen)
2259 (__psunsigned_t)context->dupcnt,
2260 (__psunsigned_t)context->flags,
2261 (__psunsigned_t)INT_GET(btree->hashval, ARCH_CONVERT),
2262 (__psunsigned_t)INT_GET(btree->before, ARCH_CONVERT),
2263 (__psunsigned_t)NULL);
2267 * Add a trace buffer entry for a context structure and a leaf block.
2270 xfs_attr_trace_l_cl(char *where, struct xfs_attr_list_context *context,
2271 struct xfs_attr_leafblock *leaf)
2273 xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CL, where,
2274 (__psunsigned_t)context->dp,
2275 (__psunsigned_t)context->cursor->hashval,
2276 (__psunsigned_t)context->cursor->blkno,
2277 (__psunsigned_t)context->cursor->offset,
2278 (__psunsigned_t)context->alist,
2279 (__psunsigned_t)context->bufsize,
2280 (__psunsigned_t)context->count,
2281 (__psunsigned_t)context->firstu,
2283 ((context->count > 0) &&
2284 !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2285 ? (ATTR_ENTRY(context->alist,
2286 context->count-1)->a_valuelen)
2288 (__psunsigned_t)context->dupcnt,
2289 (__psunsigned_t)context->flags,
2290 (__psunsigned_t)INT_GET(leaf->hdr.count, ARCH_CONVERT),
2291 (__psunsigned_t)INT_GET(leaf->entries[0].hashval, ARCH_CONVERT),
2292 (__psunsigned_t)INT_GET(leaf->entries[INT_GET(leaf->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2296 * Add a trace buffer entry for the arguments given to the routine,
2300 xfs_attr_trace_enter(int type, char *where,
2301 __psunsigned_t a2, __psunsigned_t a3,
2302 __psunsigned_t a4, __psunsigned_t a5,
2303 __psunsigned_t a6, __psunsigned_t a7,
2304 __psunsigned_t a8, __psunsigned_t a9,
2305 __psunsigned_t a10, __psunsigned_t a11,
2306 __psunsigned_t a12, __psunsigned_t a13,
2307 __psunsigned_t a14, __psunsigned_t a15)
2309 ASSERT(xfs_attr_trace_buf);
2310 ktrace_enter(xfs_attr_trace_buf, (void *)((__psunsigned_t)type),
2312 (void *)a2, (void *)a3, (void *)a4,
2313 (void *)a5, (void *)a6, (void *)a7,
2314 (void *)a8, (void *)a9, (void *)a10,
2315 (void *)a11, (void *)a12, (void *)a13,
2316 (void *)a14, (void *)a15);
2318 #endif /* XFS_ATTR_TRACE */
2321 /*========================================================================
2322 * System (pseudo) namespace attribute interface routines.
2323 *========================================================================*/
2326 posix_acl_access_set(
2327 vnode_t *vp, char *name, void *data, size_t size, int xflags)
2329 return xfs_acl_vset(vp, data, size, _ACL_TYPE_ACCESS);
2333 posix_acl_access_remove(
2334 struct vnode *vp, char *name, int xflags)
2336 return xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
2340 posix_acl_access_get(
2341 vnode_t *vp, char *name, void *data, size_t size, int xflags)
2343 return xfs_acl_vget(vp, data, size, _ACL_TYPE_ACCESS);
2347 posix_acl_access_exists(
2350 return xfs_acl_vhasacl_access(vp);
2354 posix_acl_default_set(
2355 vnode_t *vp, char *name, void *data, size_t size, int xflags)
2357 return xfs_acl_vset(vp, data, size, _ACL_TYPE_DEFAULT);
2361 posix_acl_default_get(
2362 vnode_t *vp, char *name, void *data, size_t size, int xflags)
2364 return xfs_acl_vget(vp, data, size, _ACL_TYPE_DEFAULT);
2368 posix_acl_default_remove(
2369 struct vnode *vp, char *name, int xflags)
2371 return xfs_acl_vremove(vp, _ACL_TYPE_DEFAULT);
2375 posix_acl_default_exists(
2378 return xfs_acl_vhasacl_default(vp);
2381 STATIC struct attrnames posix_acl_access = {
2382 .attr_name = "posix_acl_access",
2383 .attr_namelen = sizeof("posix_acl_access") - 1,
2384 .attr_get = posix_acl_access_get,
2385 .attr_set = posix_acl_access_set,
2386 .attr_remove = posix_acl_access_remove,
2387 .attr_exists = posix_acl_access_exists,
2390 STATIC struct attrnames posix_acl_default = {
2391 .attr_name = "posix_acl_default",
2392 .attr_namelen = sizeof("posix_acl_default") - 1,
2393 .attr_get = posix_acl_default_get,
2394 .attr_set = posix_acl_default_set,
2395 .attr_remove = posix_acl_default_remove,
2396 .attr_exists = posix_acl_default_exists,
2399 STATIC struct attrnames *attr_system_names[] =
2400 { &posix_acl_access, &posix_acl_default };
2403 /*========================================================================
2404 * Namespace-prefix-style attribute name interface routines.
2405 *========================================================================*/
2409 struct vnode *vp, char *name, void *data, size_t size, int xflags)
2413 VOP_ATTR_SET(vp, name, data, size, xflags, NULL, error);
2419 struct vnode *vp, char *name, void *data, size_t size, int xflags)
2421 int error, asize = size;
2423 VOP_ATTR_GET(vp, name, data, &asize, xflags, NULL, error);
2430 attr_generic_remove(
2431 struct vnode *vp, char *name, int xflags)
2435 VOP_ATTR_REMOVE(vp, name, xflags, NULL, error);
2440 attr_generic_listadd(
2441 attrnames_t *prefix,
2442 attrnames_t *namesp,
2447 char *p = data + *result;
2449 *result += prefix->attr_namelen;
2450 *result += namesp->attr_namelen + 1;
2455 strcpy(p, prefix->attr_name);
2456 p += prefix->attr_namelen;
2457 strcpy(p, namesp->attr_name);
2458 p += namesp->attr_namelen + 1;
2469 attrnames_t *namesp;
2472 for (i = 0; i < ATTR_SYSCOUNT; i++) {
2473 namesp = attr_system_names[i];
2474 if (!namesp->attr_exists || !namesp->attr_exists(vp))
2476 error = attr_generic_listadd(&attr_system, namesp,
2477 data, size, result);
2486 struct vnode *vp, void *data, size_t size, int xflags, ssize_t *result)
2488 attrlist_cursor_kern_t cursor = { 0 };
2491 VOP_ATTR_LIST(vp, data, size, xflags, &cursor, NULL, error);
2495 return attr_system_list(vp, data, size, result);
2499 attr_lookup_namespace(
2501 struct attrnames **names,
2506 for (i = 0; i < nnames; i++)
2507 if (!strncmp(name, names[i]->attr_name, names[i]->attr_namelen))
2513 * Some checks to prevent people abusing EAs to get over quota:
2514 * - Don't allow modifying user EAs on devices/symlinks;
2515 * - Don't allow modifying user EAs if sticky bit set;
2522 struct inode *inode = LINVFS_GET_IP(vp);
2524 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2526 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
2527 !capable(CAP_SYS_ADMIN))
2529 if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
2530 (current_fsuid(cred) != inode->i_uid) && !capable(CAP_FOWNER))
2536 attr_trusted_capable(
2540 struct inode *inode = LINVFS_GET_IP(vp);
2542 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2544 if (!capable(CAP_SYS_ADMIN))
2550 attr_secure_capable(
2554 return -ENOSECURITY;
2559 struct vnode *vp, char *name, void *data, size_t size, int xflags)
2561 attrnames_t *namesp;
2564 if (xflags & ATTR_CREATE)
2567 namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2570 error = namesp->attr_set(vp, name, data, size, xflags);
2572 error = vn_revalidate(vp);
2578 struct vnode *vp, char *name, void *data, size_t size, int xflags)
2580 attrnames_t *namesp;
2582 namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2585 return namesp->attr_get(vp, name, data, size, xflags);
2590 struct vnode *vp, char *name, int xflags)
2592 attrnames_t *namesp;
2594 namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2597 return namesp->attr_remove(vp, name, xflags);
2600 struct attrnames attr_system = {
2601 .attr_name = "system.",
2602 .attr_namelen = sizeof("system.") - 1,
2603 .attr_flag = ATTR_SYSTEM,
2604 .attr_get = attr_system_get,
2605 .attr_set = attr_system_set,
2606 .attr_remove = attr_system_remove,
2607 .attr_capable = (attrcapable_t)fs_noerr,
2610 struct attrnames attr_trusted = {
2611 .attr_name = "trusted.",
2612 .attr_namelen = sizeof("trusted.") - 1,
2613 .attr_flag = ATTR_ROOT,
2614 .attr_get = attr_generic_get,
2615 .attr_set = attr_generic_set,
2616 .attr_remove = attr_generic_remove,
2617 .attr_capable = attr_trusted_capable,
2620 struct attrnames attr_secure = {
2621 .attr_name = "security.",
2622 .attr_namelen = sizeof("security.") - 1,
2623 .attr_flag = ATTR_SECURE,
2624 .attr_get = attr_generic_get,
2625 .attr_set = attr_generic_set,
2626 .attr_remove = attr_generic_remove,
2627 .attr_capable = attr_secure_capable,
2630 struct attrnames attr_user = {
2631 .attr_name = "user.",
2632 .attr_namelen = sizeof("user.") - 1,
2633 .attr_get = attr_generic_get,
2634 .attr_set = attr_generic_set,
2635 .attr_remove = attr_generic_remove,
2636 .attr_capable = attr_user_capable,
2639 struct attrnames *attr_namespaces[] =
2640 { &attr_system, &attr_trusted, &attr_secure, &attr_user };