ext4: Support large files
[linux-2.6] / fs / xfs / xfs_attr.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
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.
8  *
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.
13  *
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
17  */
18
19 #include <linux/capability.h>
20
21 #include "xfs.h"
22 #include "xfs_fs.h"
23 #include "xfs_types.h"
24 #include "xfs_bit.h"
25 #include "xfs_log.h"
26 #include "xfs_inum.h"
27 #include "xfs_trans.h"
28 #include "xfs_sb.h"
29 #include "xfs_ag.h"
30 #include "xfs_dir2.h"
31 #include "xfs_dmapi.h"
32 #include "xfs_mount.h"
33 #include "xfs_da_btree.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_alloc.h"
42 #include "xfs_btree.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_bmap.h"
45 #include "xfs_attr.h"
46 #include "xfs_attr_leaf.h"
47 #include "xfs_error.h"
48 #include "xfs_quota.h"
49 #include "xfs_trans_space.h"
50 #include "xfs_acl.h"
51 #include "xfs_rw.h"
52 #include "xfs_vnodeops.h"
53
54 /*
55  * xfs_attr.c
56  *
57  * Provide the external interfaces to manage attribute lists.
58  */
59
60 #define ATTR_SYSCOUNT   2
61 static struct attrnames posix_acl_access;
62 static struct attrnames posix_acl_default;
63 static struct attrnames *attr_system_names[ATTR_SYSCOUNT];
64
65 /*========================================================================
66  * Function prototypes for the kernel.
67  *========================================================================*/
68
69 /*
70  * Internal routines when attribute list fits inside the inode.
71  */
72 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
73
74 /*
75  * Internal routines when attribute list is one block.
76  */
77 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
78 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
79 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
80 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
81
82 /*
83  * Internal routines when attribute list is more than one block.
84  */
85 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
86 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
87 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
88 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
89 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
90 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
91
92 /*
93  * Routines to manipulate out-of-line attribute values.
94  */
95 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
96 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
97
98 #define ATTR_RMTVALUE_MAPSIZE   1       /* # of map entries at once */
99
100 #if defined(XFS_ATTR_TRACE)
101 ktrace_t *xfs_attr_trace_buf;
102 #endif
103
104
105 /*========================================================================
106  * Overall external interface routines.
107  *========================================================================*/
108
109 int
110 xfs_attr_fetch(xfs_inode_t *ip, const char *name, int namelen,
111                char *value, int *valuelenp, int flags, struct cred *cred)
112 {
113         xfs_da_args_t   args;
114         int             error;
115
116         if ((XFS_IFORK_Q(ip) == 0) ||
117             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
118              ip->i_d.di_anextents == 0))
119                 return(ENOATTR);
120
121         /*
122          * Fill in the arg structure for this request.
123          */
124         memset((char *)&args, 0, sizeof(args));
125         args.name = name;
126         args.namelen = namelen;
127         args.value = value;
128         args.valuelen = *valuelenp;
129         args.flags = flags;
130         args.hashval = xfs_da_hashname(args.name, args.namelen);
131         args.dp = ip;
132         args.whichfork = XFS_ATTR_FORK;
133
134         /*
135          * Decide on what work routines to call based on the inode size.
136          */
137         if (XFS_IFORK_Q(ip) == 0 ||
138             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
139              ip->i_d.di_anextents == 0)) {
140                 error = XFS_ERROR(ENOATTR);
141         } else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
142                 error = xfs_attr_shortform_getvalue(&args);
143         } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
144                 error = xfs_attr_leaf_get(&args);
145         } else {
146                 error = xfs_attr_node_get(&args);
147         }
148
149         /*
150          * Return the number of bytes in the value to the caller.
151          */
152         *valuelenp = args.valuelen;
153
154         if (error == EEXIST)
155                 error = 0;
156         return(error);
157 }
158
159 int
160 xfs_attr_get(
161         xfs_inode_t     *ip,
162         const char      *name,
163         char            *value,
164         int             *valuelenp,
165         int             flags,
166         cred_t          *cred)
167 {
168         int             error, namelen;
169
170         XFS_STATS_INC(xs_attr_get);
171
172         if (!name)
173                 return(EINVAL);
174         namelen = strlen(name);
175         if (namelen >= MAXNAMELEN)
176                 return(EFAULT);         /* match IRIX behaviour */
177
178         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
179                 return(EIO);
180
181         xfs_ilock(ip, XFS_ILOCK_SHARED);
182         error = xfs_attr_fetch(ip, name, namelen, value, valuelenp, flags, cred);
183         xfs_iunlock(ip, XFS_ILOCK_SHARED);
184         return(error);
185 }
186
187 int
188 xfs_attr_set_int(xfs_inode_t *dp, const char *name, int namelen,
189                  char *value, int valuelen, int flags)
190 {
191         xfs_da_args_t   args;
192         xfs_fsblock_t   firstblock;
193         xfs_bmap_free_t flist;
194         int             error, err2, committed;
195         int             local, size;
196         uint            nblks;
197         xfs_mount_t     *mp = dp->i_mount;
198         int             rsvd = (flags & ATTR_ROOT) != 0;
199
200         /*
201          * Attach the dquots to the inode.
202          */
203         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
204                 return (error);
205
206         /*
207          * If the inode doesn't have an attribute fork, add one.
208          * (inode must not be locked when we call this routine)
209          */
210         if (XFS_IFORK_Q(dp) == 0) {
211                 int sf_size = sizeof(xfs_attr_sf_hdr_t) +
212                               XFS_ATTR_SF_ENTSIZE_BYNAME(namelen, valuelen);
213
214                 if ((error = xfs_bmap_add_attrfork(dp, sf_size, rsvd)))
215                         return(error);
216         }
217
218         /*
219          * Fill in the arg structure for this request.
220          */
221         memset((char *)&args, 0, sizeof(args));
222         args.name = name;
223         args.namelen = namelen;
224         args.value = value;
225         args.valuelen = valuelen;
226         args.flags = flags;
227         args.hashval = xfs_da_hashname(args.name, args.namelen);
228         args.dp = dp;
229         args.firstblock = &firstblock;
230         args.flist = &flist;
231         args.whichfork = XFS_ATTR_FORK;
232         args.addname = 1;
233         args.oknoent = 1;
234
235         /*
236          * Determine space new attribute will use, and if it would be
237          * "local" or "remote" (note: local != inline).
238          */
239         size = xfs_attr_leaf_newentsize(namelen, valuelen,
240                                         mp->m_sb.sb_blocksize, &local);
241
242         nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
243         if (local) {
244                 if (size > (mp->m_sb.sb_blocksize >> 1)) {
245                         /* Double split possible */
246                         nblks <<= 1;
247                 }
248         } else {
249                 uint    dblocks = XFS_B_TO_FSB(mp, valuelen);
250                 /* Out of line attribute, cannot double split, but make
251                  * room for the attribute value itself.
252                  */
253                 nblks += dblocks;
254                 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
255         }
256
257         /* Size is now blocks for attribute data */
258         args.total = nblks;
259
260         /*
261          * Start our first transaction of the day.
262          *
263          * All future transactions during this code must be "chained" off
264          * this one via the trans_dup() call.  All transactions will contain
265          * the inode, and the inode will always be marked with trans_ihold().
266          * Since the inode will be locked in all transactions, we must log
267          * the inode in every transaction to let it float upward through
268          * the log.
269          */
270         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
271
272         /*
273          * Root fork attributes can use reserved data blocks for this
274          * operation if necessary
275          */
276
277         if (rsvd)
278                 args.trans->t_flags |= XFS_TRANS_RESERVE;
279
280         if ((error = xfs_trans_reserve(args.trans, (uint) nblks,
281                                       XFS_ATTRSET_LOG_RES(mp, nblks),
282                                       0, XFS_TRANS_PERM_LOG_RES,
283                                       XFS_ATTRSET_LOG_COUNT))) {
284                 xfs_trans_cancel(args.trans, 0);
285                 return(error);
286         }
287         xfs_ilock(dp, XFS_ILOCK_EXCL);
288
289         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, args.trans, dp, nblks, 0,
290                          rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
291                                 XFS_QMOPT_RES_REGBLKS);
292         if (error) {
293                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
294                 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
295                 return (error);
296         }
297
298         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
299         xfs_trans_ihold(args.trans, dp);
300
301         /*
302          * If the attribute list is non-existent or a shortform list,
303          * upgrade it to a single-leaf-block attribute list.
304          */
305         if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
306             ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
307              (dp->i_d.di_anextents == 0))) {
308
309                 /*
310                  * Build initial attribute list (if required).
311                  */
312                 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
313                         xfs_attr_shortform_create(&args);
314
315                 /*
316                  * Try to add the attr to the attribute list in
317                  * the inode.
318                  */
319                 error = xfs_attr_shortform_addname(&args);
320                 if (error != ENOSPC) {
321                         /*
322                          * Commit the shortform mods, and we're done.
323                          * NOTE: this is also the error path (EEXIST, etc).
324                          */
325                         ASSERT(args.trans != NULL);
326
327                         /*
328                          * If this is a synchronous mount, make sure that
329                          * the transaction goes to disk before returning
330                          * to the user.
331                          */
332                         if (mp->m_flags & XFS_MOUNT_WSYNC) {
333                                 xfs_trans_set_sync(args.trans);
334                         }
335                         err2 = xfs_trans_commit(args.trans,
336                                                  XFS_TRANS_RELEASE_LOG_RES);
337                         xfs_iunlock(dp, XFS_ILOCK_EXCL);
338
339                         /*
340                          * Hit the inode change time.
341                          */
342                         if (!error && (flags & ATTR_KERNOTIME) == 0) {
343                                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
344                         }
345                         return(error == 0 ? err2 : error);
346                 }
347
348                 /*
349                  * It won't fit in the shortform, transform to a leaf block.
350                  * GROT: another possible req'mt for a double-split btree op.
351                  */
352                 XFS_BMAP_INIT(args.flist, args.firstblock);
353                 error = xfs_attr_shortform_to_leaf(&args);
354                 if (!error) {
355                         error = xfs_bmap_finish(&args.trans, args.flist,
356                                                 &committed);
357                 }
358                 if (error) {
359                         ASSERT(committed);
360                         args.trans = NULL;
361                         xfs_bmap_cancel(&flist);
362                         goto out;
363                 }
364
365                 /*
366                  * bmap_finish() may have committed the last trans and started
367                  * a new one.  We need the inode to be in all transactions.
368                  */
369                 if (committed) {
370                         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
371                         xfs_trans_ihold(args.trans, dp);
372                 }
373
374                 /*
375                  * Commit the leaf transformation.  We'll need another (linked)
376                  * transaction to add the new attribute to the leaf.
377                  */
378                 if ((error = xfs_attr_rolltrans(&args.trans, dp)))
379                         goto out;
380
381         }
382
383         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
384                 error = xfs_attr_leaf_addname(&args);
385         } else {
386                 error = xfs_attr_node_addname(&args);
387         }
388         if (error) {
389                 goto out;
390         }
391
392         /*
393          * If this is a synchronous mount, make sure that the
394          * transaction goes to disk before returning to the user.
395          */
396         if (mp->m_flags & XFS_MOUNT_WSYNC) {
397                 xfs_trans_set_sync(args.trans);
398         }
399
400         /*
401          * Commit the last in the sequence of transactions.
402          */
403         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
404         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
405         xfs_iunlock(dp, XFS_ILOCK_EXCL);
406
407         /*
408          * Hit the inode change time.
409          */
410         if (!error && (flags & ATTR_KERNOTIME) == 0) {
411                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
412         }
413
414         return(error);
415
416 out:
417         if (args.trans)
418                 xfs_trans_cancel(args.trans,
419                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
420         xfs_iunlock(dp, XFS_ILOCK_EXCL);
421         return(error);
422 }
423
424 int
425 xfs_attr_set(
426         xfs_inode_t     *dp,
427         const char      *name,
428         char            *value,
429         int             valuelen,
430         int             flags)
431 {
432         int             namelen;
433
434         namelen = strlen(name);
435         if (namelen >= MAXNAMELEN)
436                 return EFAULT;          /* match IRIX behaviour */
437
438         XFS_STATS_INC(xs_attr_set);
439
440         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
441                 return (EIO);
442
443         return xfs_attr_set_int(dp, name, namelen, value, valuelen, flags);
444 }
445
446 /*
447  * Generic handler routine to remove a name from an attribute list.
448  * Transitions attribute list from Btree to shortform as necessary.
449  */
450 int
451 xfs_attr_remove_int(xfs_inode_t *dp, const char *name, int namelen, int flags)
452 {
453         xfs_da_args_t   args;
454         xfs_fsblock_t   firstblock;
455         xfs_bmap_free_t flist;
456         int             error;
457         xfs_mount_t     *mp = dp->i_mount;
458
459         /*
460          * Fill in the arg structure for this request.
461          */
462         memset((char *)&args, 0, sizeof(args));
463         args.name = name;
464         args.namelen = namelen;
465         args.flags = flags;
466         args.hashval = xfs_da_hashname(args.name, args.namelen);
467         args.dp = dp;
468         args.firstblock = &firstblock;
469         args.flist = &flist;
470         args.total = 0;
471         args.whichfork = XFS_ATTR_FORK;
472
473         /*
474          * Attach the dquots to the inode.
475          */
476         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
477                 return (error);
478
479         /*
480          * Start our first transaction of the day.
481          *
482          * All future transactions during this code must be "chained" off
483          * this one via the trans_dup() call.  All transactions will contain
484          * the inode, and the inode will always be marked with trans_ihold().
485          * Since the inode will be locked in all transactions, we must log
486          * the inode in every transaction to let it float upward through
487          * the log.
488          */
489         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
490
491         /*
492          * Root fork attributes can use reserved data blocks for this
493          * operation if necessary
494          */
495
496         if (flags & ATTR_ROOT)
497                 args.trans->t_flags |= XFS_TRANS_RESERVE;
498
499         if ((error = xfs_trans_reserve(args.trans,
500                                       XFS_ATTRRM_SPACE_RES(mp),
501                                       XFS_ATTRRM_LOG_RES(mp),
502                                       0, XFS_TRANS_PERM_LOG_RES,
503                                       XFS_ATTRRM_LOG_COUNT))) {
504                 xfs_trans_cancel(args.trans, 0);
505                 return(error);
506         }
507
508         xfs_ilock(dp, XFS_ILOCK_EXCL);
509         /*
510          * No need to make quota reservations here. We expect to release some
511          * blocks not allocate in the common case.
512          */
513         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
514         xfs_trans_ihold(args.trans, dp);
515
516         /*
517          * Decide on what work routines to call based on the inode size.
518          */
519         if (XFS_IFORK_Q(dp) == 0 ||
520             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
521              dp->i_d.di_anextents == 0)) {
522                 error = XFS_ERROR(ENOATTR);
523                 goto out;
524         }
525         if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
526                 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
527                 error = xfs_attr_shortform_remove(&args);
528                 if (error) {
529                         goto out;
530                 }
531         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
532                 error = xfs_attr_leaf_removename(&args);
533         } else {
534                 error = xfs_attr_node_removename(&args);
535         }
536         if (error) {
537                 goto out;
538         }
539
540         /*
541          * If this is a synchronous mount, make sure that the
542          * transaction goes to disk before returning to the user.
543          */
544         if (mp->m_flags & XFS_MOUNT_WSYNC) {
545                 xfs_trans_set_sync(args.trans);
546         }
547
548         /*
549          * Commit the last in the sequence of transactions.
550          */
551         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
552         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
553         xfs_iunlock(dp, XFS_ILOCK_EXCL);
554
555         /*
556          * Hit the inode change time.
557          */
558         if (!error && (flags & ATTR_KERNOTIME) == 0) {
559                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
560         }
561
562         return(error);
563
564 out:
565         if (args.trans)
566                 xfs_trans_cancel(args.trans,
567                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
568         xfs_iunlock(dp, XFS_ILOCK_EXCL);
569         return(error);
570 }
571
572 int
573 xfs_attr_remove(
574         xfs_inode_t     *dp,
575         const char      *name,
576         int             flags)
577 {
578         int             namelen;
579
580         namelen = strlen(name);
581         if (namelen >= MAXNAMELEN)
582                 return EFAULT;          /* match IRIX behaviour */
583
584         XFS_STATS_INC(xs_attr_remove);
585
586         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
587                 return (EIO);
588
589         xfs_ilock(dp, XFS_ILOCK_SHARED);
590         if (XFS_IFORK_Q(dp) == 0 ||
591                    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
592                     dp->i_d.di_anextents == 0)) {
593                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
594                 return(XFS_ERROR(ENOATTR));
595         }
596         xfs_iunlock(dp, XFS_ILOCK_SHARED);
597
598         return xfs_attr_remove_int(dp, name, namelen, flags);
599 }
600
601 int                                                             /* error */
602 xfs_attr_list_int(xfs_attr_list_context_t *context)
603 {
604         int error;
605         xfs_inode_t *dp = context->dp;
606
607         /*
608          * Decide on what work routines to call based on the inode size.
609          */
610         if (XFS_IFORK_Q(dp) == 0 ||
611             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
612              dp->i_d.di_anextents == 0)) {
613                 error = 0;
614         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
615                 error = xfs_attr_shortform_list(context);
616         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
617                 error = xfs_attr_leaf_list(context);
618         } else {
619                 error = xfs_attr_node_list(context);
620         }
621         return error;
622 }
623
624 #define ATTR_ENTBASESIZE                /* minimum bytes used by an attr */ \
625         (((struct attrlist_ent *) 0)->a_name - (char *) 0)
626 #define ATTR_ENTSIZE(namelen)           /* actual bytes used by an attr */ \
627         ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
628          & ~(sizeof(u_int32_t)-1))
629
630 /*
631  * Format an attribute and copy it out to the user's buffer.
632  * Take care to check values and protect against them changing later,
633  * we may be reading them directly out of a user buffer.
634  */
635 /*ARGSUSED*/
636 STATIC int
637 xfs_attr_put_listent(xfs_attr_list_context_t *context, attrnames_t *namesp,
638                      char *name, int namelen,
639                      int valuelen, char *value)
640 {
641         attrlist_ent_t *aep;
642         int arraytop;
643
644         ASSERT(!(context->flags & ATTR_KERNOVAL));
645         ASSERT(context->count >= 0);
646         ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
647         ASSERT(context->firstu >= sizeof(*context->alist));
648         ASSERT(context->firstu <= context->bufsize);
649
650         arraytop = sizeof(*context->alist) +
651                         context->count * sizeof(context->alist->al_offset[0]);
652         context->firstu -= ATTR_ENTSIZE(namelen);
653         if (context->firstu < arraytop) {
654                 xfs_attr_trace_l_c("buffer full", context);
655                 context->alist->al_more = 1;
656                 context->seen_enough = 1;
657                 return 1;
658         }
659
660         aep = (attrlist_ent_t *)&(((char *)context->alist)[ context->firstu ]);
661         aep->a_valuelen = valuelen;
662         memcpy(aep->a_name, name, namelen);
663         aep->a_name[ namelen ] = 0;
664         context->alist->al_offset[ context->count++ ] = context->firstu;
665         context->alist->al_count = context->count;
666         xfs_attr_trace_l_c("add", context);
667         return 0;
668 }
669
670 STATIC int
671 xfs_attr_kern_list(xfs_attr_list_context_t *context, attrnames_t *namesp,
672                      char *name, int namelen,
673                      int valuelen, char *value)
674 {
675         char *offset;
676         int arraytop;
677
678         ASSERT(context->count >= 0);
679
680         arraytop = context->count + namesp->attr_namelen + namelen + 1;
681         if (arraytop > context->firstu) {
682                 context->count = -1;    /* insufficient space */
683                 return 1;
684         }
685         offset = (char *)context->alist + context->count;
686         strncpy(offset, namesp->attr_name, namesp->attr_namelen);
687         offset += namesp->attr_namelen;
688         strncpy(offset, name, namelen);                 /* real name */
689         offset += namelen;
690         *offset = '\0';
691         context->count += namesp->attr_namelen + namelen + 1;
692         return 0;
693 }
694
695 /*ARGSUSED*/
696 STATIC int
697 xfs_attr_kern_list_sizes(xfs_attr_list_context_t *context, attrnames_t *namesp,
698                      char *name, int namelen,
699                      int valuelen, char *value)
700 {
701         context->count += namesp->attr_namelen + namelen + 1;
702         return 0;
703 }
704
705 /*
706  * Generate a list of extended attribute names and optionally
707  * also value lengths.  Positive return value follows the XFS
708  * convention of being an error, zero or negative return code
709  * is the length of the buffer returned (negated), indicating
710  * success.
711  */
712 int
713 xfs_attr_list(
714         xfs_inode_t     *dp,
715         char            *buffer,
716         int             bufsize,
717         int             flags,
718         attrlist_cursor_kern_t *cursor)
719 {
720         xfs_attr_list_context_t context;
721         int error;
722
723         XFS_STATS_INC(xs_attr_list);
724
725         /*
726          * Validate the cursor.
727          */
728         if (cursor->pad1 || cursor->pad2)
729                 return(XFS_ERROR(EINVAL));
730         if ((cursor->initted == 0) &&
731             (cursor->hashval || cursor->blkno || cursor->offset))
732                 return XFS_ERROR(EINVAL);
733
734         /*
735          * Check for a properly aligned buffer.
736          */
737         if (((long)buffer) & (sizeof(int)-1))
738                 return XFS_ERROR(EFAULT);
739         if (flags & ATTR_KERNOVAL)
740                 bufsize = 0;
741
742         /*
743          * Initialize the output buffer.
744          */
745         context.dp = dp;
746         context.cursor = cursor;
747         context.count = 0;
748         context.dupcnt = 0;
749         context.resynch = 1;
750         context.flags = flags;
751         context.seen_enough = 0;
752         context.alist = (attrlist_t *)buffer;
753         context.put_value = 0;
754
755         if (flags & ATTR_KERNAMELS) {
756                 context.bufsize = bufsize;
757                 context.firstu = context.bufsize;
758                 if (flags & ATTR_KERNOVAL)
759                         context.put_listent = xfs_attr_kern_list_sizes;
760                 else
761                         context.put_listent = xfs_attr_kern_list;
762         } else {
763                 context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
764                 context.firstu = context.bufsize;
765                 context.alist->al_count = 0;
766                 context.alist->al_more = 0;
767                 context.alist->al_offset[0] = context.bufsize;
768                 context.put_listent = xfs_attr_put_listent;
769         }
770
771         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
772                 return EIO;
773
774         xfs_ilock(dp, XFS_ILOCK_SHARED);
775         xfs_attr_trace_l_c("syscall start", &context);
776
777         error = xfs_attr_list_int(&context);
778
779         xfs_iunlock(dp, XFS_ILOCK_SHARED);
780         xfs_attr_trace_l_c("syscall end", &context);
781
782         if (context.flags & (ATTR_KERNOVAL|ATTR_KERNAMELS)) {
783                 /* must return negated buffer size or the error */
784                 if (context.count < 0)
785                         error = XFS_ERROR(ERANGE);
786                 else
787                         error = -context.count;
788         } else
789                 ASSERT(error >= 0);
790
791         return error;
792 }
793
794 int                                                             /* error */
795 xfs_attr_inactive(xfs_inode_t *dp)
796 {
797         xfs_trans_t *trans;
798         xfs_mount_t *mp;
799         int error;
800
801         mp = dp->i_mount;
802         ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
803
804         xfs_ilock(dp, XFS_ILOCK_SHARED);
805         if ((XFS_IFORK_Q(dp) == 0) ||
806             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
807             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
808              dp->i_d.di_anextents == 0)) {
809                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
810                 return(0);
811         }
812         xfs_iunlock(dp, XFS_ILOCK_SHARED);
813
814         /*
815          * Start our first transaction of the day.
816          *
817          * All future transactions during this code must be "chained" off
818          * this one via the trans_dup() call.  All transactions will contain
819          * the inode, and the inode will always be marked with trans_ihold().
820          * Since the inode will be locked in all transactions, we must log
821          * the inode in every transaction to let it float upward through
822          * the log.
823          */
824         trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
825         if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
826                                       XFS_TRANS_PERM_LOG_RES,
827                                       XFS_ATTRINVAL_LOG_COUNT))) {
828                 xfs_trans_cancel(trans, 0);
829                 return(error);
830         }
831         xfs_ilock(dp, XFS_ILOCK_EXCL);
832
833         /*
834          * No need to make quota reservations here. We expect to release some
835          * blocks, not allocate, in the common case.
836          */
837         xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
838         xfs_trans_ihold(trans, dp);
839
840         /*
841          * Decide on what work routines to call based on the inode size.
842          */
843         if ((XFS_IFORK_Q(dp) == 0) ||
844             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
845             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
846              dp->i_d.di_anextents == 0)) {
847                 error = 0;
848                 goto out;
849         }
850         error = xfs_attr_root_inactive(&trans, dp);
851         if (error)
852                 goto out;
853         /*
854          * signal synchronous inactive transactions unless this
855          * is a synchronous mount filesystem in which case we
856          * know that we're here because we've been called out of
857          * xfs_inactive which means that the last reference is gone
858          * and the unlink transaction has already hit the disk so
859          * async inactive transactions are safe.
860          */
861         if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
862                                 (!(mp->m_flags & XFS_MOUNT_WSYNC)
863                                  ? 1 : 0))))
864                 goto out;
865
866         /*
867          * Commit the last in the sequence of transactions.
868          */
869         xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
870         error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
871         xfs_iunlock(dp, XFS_ILOCK_EXCL);
872
873         return(error);
874
875 out:
876         xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
877         xfs_iunlock(dp, XFS_ILOCK_EXCL);
878         return(error);
879 }
880
881
882
883 /*========================================================================
884  * External routines when attribute list is inside the inode
885  *========================================================================*/
886
887 /*
888  * Add a name to the shortform attribute list structure
889  * This is the external routine.
890  */
891 STATIC int
892 xfs_attr_shortform_addname(xfs_da_args_t *args)
893 {
894         int newsize, forkoff, retval;
895
896         retval = xfs_attr_shortform_lookup(args);
897         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
898                 return(retval);
899         } else if (retval == EEXIST) {
900                 if (args->flags & ATTR_CREATE)
901                         return(retval);
902                 retval = xfs_attr_shortform_remove(args);
903                 ASSERT(retval == 0);
904         }
905
906         if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
907             args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
908                 return(XFS_ERROR(ENOSPC));
909
910         newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
911         newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
912
913         forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
914         if (!forkoff)
915                 return(XFS_ERROR(ENOSPC));
916
917         xfs_attr_shortform_add(args, forkoff);
918         return(0);
919 }
920
921
922 /*========================================================================
923  * External routines when attribute list is one block
924  *========================================================================*/
925
926 /*
927  * Add a name to the leaf attribute list structure
928  *
929  * This leaf block cannot have a "remote" value, we only call this routine
930  * if bmap_one_block() says there is only one block (ie: no remote blks).
931  */
932 int
933 xfs_attr_leaf_addname(xfs_da_args_t *args)
934 {
935         xfs_inode_t *dp;
936         xfs_dabuf_t *bp;
937         int retval, error, committed, forkoff;
938
939         /*
940          * Read the (only) block in the attribute list in.
941          */
942         dp = args->dp;
943         args->blkno = 0;
944         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
945                                              XFS_ATTR_FORK);
946         if (error)
947                 return(error);
948         ASSERT(bp != NULL);
949
950         /*
951          * Look up the given attribute in the leaf block.  Figure out if
952          * the given flags produce an error or call for an atomic rename.
953          */
954         retval = xfs_attr_leaf_lookup_int(bp, args);
955         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
956                 xfs_da_brelse(args->trans, bp);
957                 return(retval);
958         } else if (retval == EEXIST) {
959                 if (args->flags & ATTR_CREATE) {        /* pure create op */
960                         xfs_da_brelse(args->trans, bp);
961                         return(retval);
962                 }
963                 args->rename = 1;                       /* an atomic rename */
964                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
965                 args->index2 = args->index;
966                 args->rmtblkno2 = args->rmtblkno;
967                 args->rmtblkcnt2 = args->rmtblkcnt;
968         }
969
970         /*
971          * Add the attribute to the leaf block, transitioning to a Btree
972          * if required.
973          */
974         retval = xfs_attr_leaf_add(bp, args);
975         xfs_da_buf_done(bp);
976         if (retval == ENOSPC) {
977                 /*
978                  * Promote the attribute list to the Btree format, then
979                  * Commit that transaction so that the node_addname() call
980                  * can manage its own transactions.
981                  */
982                 XFS_BMAP_INIT(args->flist, args->firstblock);
983                 error = xfs_attr_leaf_to_node(args);
984                 if (!error) {
985                         error = xfs_bmap_finish(&args->trans, args->flist,
986                                                 &committed);
987                 }
988                 if (error) {
989                         ASSERT(committed);
990                         args->trans = NULL;
991                         xfs_bmap_cancel(args->flist);
992                         return(error);
993                 }
994
995                 /*
996                  * bmap_finish() may have committed the last trans and started
997                  * a new one.  We need the inode to be in all transactions.
998                  */
999                 if (committed) {
1000                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1001                         xfs_trans_ihold(args->trans, dp);
1002                 }
1003
1004                 /*
1005                  * Commit the current trans (including the inode) and start
1006                  * a new one.
1007                  */
1008                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1009                         return (error);
1010
1011                 /*
1012                  * Fob the whole rest of the problem off on the Btree code.
1013                  */
1014                 error = xfs_attr_node_addname(args);
1015                 return(error);
1016         }
1017
1018         /*
1019          * Commit the transaction that added the attr name so that
1020          * later routines can manage their own transactions.
1021          */
1022         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1023                 return (error);
1024
1025         /*
1026          * If there was an out-of-line value, allocate the blocks we
1027          * identified for its storage and copy the value.  This is done
1028          * after we create the attribute so that we don't overflow the
1029          * maximum size of a transaction and/or hit a deadlock.
1030          */
1031         if (args->rmtblkno > 0) {
1032                 error = xfs_attr_rmtval_set(args);
1033                 if (error)
1034                         return(error);
1035         }
1036
1037         /*
1038          * If this is an atomic rename operation, we must "flip" the
1039          * incomplete flags on the "new" and "old" attribute/value pairs
1040          * so that one disappears and one appears atomically.  Then we
1041          * must remove the "old" attribute/value pair.
1042          */
1043         if (args->rename) {
1044                 /*
1045                  * In a separate transaction, set the incomplete flag on the
1046                  * "old" attr and clear the incomplete flag on the "new" attr.
1047                  */
1048                 error = xfs_attr_leaf_flipflags(args);
1049                 if (error)
1050                         return(error);
1051
1052                 /*
1053                  * Dismantle the "old" attribute/value pair by removing
1054                  * a "remote" value (if it exists).
1055                  */
1056                 args->index = args->index2;
1057                 args->blkno = args->blkno2;
1058                 args->rmtblkno = args->rmtblkno2;
1059                 args->rmtblkcnt = args->rmtblkcnt2;
1060                 if (args->rmtblkno) {
1061                         error = xfs_attr_rmtval_remove(args);
1062                         if (error)
1063                                 return(error);
1064                 }
1065
1066                 /*
1067                  * Read in the block containing the "old" attr, then
1068                  * remove the "old" attr from that block (neat, huh!)
1069                  */
1070                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1071                                                      &bp, XFS_ATTR_FORK);
1072                 if (error)
1073                         return(error);
1074                 ASSERT(bp != NULL);
1075                 (void)xfs_attr_leaf_remove(bp, args);
1076
1077                 /*
1078                  * If the result is small enough, shrink it all into the inode.
1079                  */
1080                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1081                         XFS_BMAP_INIT(args->flist, args->firstblock);
1082                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1083                         /* bp is gone due to xfs_da_shrink_inode */
1084                         if (!error) {
1085                                 error = xfs_bmap_finish(&args->trans,
1086                                                         args->flist,
1087                                                         &committed);
1088                         }
1089                         if (error) {
1090                                 ASSERT(committed);
1091                                 args->trans = NULL;
1092                                 xfs_bmap_cancel(args->flist);
1093                                 return(error);
1094                         }
1095
1096                         /*
1097                          * bmap_finish() may have committed the last trans
1098                          * and started a new one.  We need the inode to be
1099                          * in all transactions.
1100                          */
1101                         if (committed) {
1102                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1103                                 xfs_trans_ihold(args->trans, dp);
1104                         }
1105                 } else
1106                         xfs_da_buf_done(bp);
1107
1108                 /*
1109                  * Commit the remove and start the next trans in series.
1110                  */
1111                 error = xfs_attr_rolltrans(&args->trans, dp);
1112
1113         } else if (args->rmtblkno > 0) {
1114                 /*
1115                  * Added a "remote" value, just clear the incomplete flag.
1116                  */
1117                 error = xfs_attr_leaf_clearflag(args);
1118         }
1119         return(error);
1120 }
1121
1122 /*
1123  * Remove a name from the leaf attribute list structure
1124  *
1125  * This leaf block cannot have a "remote" value, we only call this routine
1126  * if bmap_one_block() says there is only one block (ie: no remote blks).
1127  */
1128 STATIC int
1129 xfs_attr_leaf_removename(xfs_da_args_t *args)
1130 {
1131         xfs_inode_t *dp;
1132         xfs_dabuf_t *bp;
1133         int error, committed, forkoff;
1134
1135         /*
1136          * Remove the attribute.
1137          */
1138         dp = args->dp;
1139         args->blkno = 0;
1140         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1141                                              XFS_ATTR_FORK);
1142         if (error) {
1143                 return(error);
1144         }
1145
1146         ASSERT(bp != NULL);
1147         error = xfs_attr_leaf_lookup_int(bp, args);
1148         if (error == ENOATTR) {
1149                 xfs_da_brelse(args->trans, bp);
1150                 return(error);
1151         }
1152
1153         (void)xfs_attr_leaf_remove(bp, args);
1154
1155         /*
1156          * If the result is small enough, shrink it all into the inode.
1157          */
1158         if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1159                 XFS_BMAP_INIT(args->flist, args->firstblock);
1160                 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1161                 /* bp is gone due to xfs_da_shrink_inode */
1162                 if (!error) {
1163                         error = xfs_bmap_finish(&args->trans, args->flist,
1164                                                 &committed);
1165                 }
1166                 if (error) {
1167                         ASSERT(committed);
1168                         args->trans = NULL;
1169                         xfs_bmap_cancel(args->flist);
1170                         return(error);
1171                 }
1172
1173                 /*
1174                  * bmap_finish() may have committed the last trans and started
1175                  * a new one.  We need the inode to be in all transactions.
1176                  */
1177                 if (committed) {
1178                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1179                         xfs_trans_ihold(args->trans, dp);
1180                 }
1181         } else
1182                 xfs_da_buf_done(bp);
1183         return(0);
1184 }
1185
1186 /*
1187  * Look up a name in a leaf attribute list structure.
1188  *
1189  * This leaf block cannot have a "remote" value, we only call this routine
1190  * if bmap_one_block() says there is only one block (ie: no remote blks).
1191  */
1192 STATIC int
1193 xfs_attr_leaf_get(xfs_da_args_t *args)
1194 {
1195         xfs_dabuf_t *bp;
1196         int error;
1197
1198         args->blkno = 0;
1199         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1200                                              XFS_ATTR_FORK);
1201         if (error)
1202                 return(error);
1203         ASSERT(bp != NULL);
1204
1205         error = xfs_attr_leaf_lookup_int(bp, args);
1206         if (error != EEXIST)  {
1207                 xfs_da_brelse(args->trans, bp);
1208                 return(error);
1209         }
1210         error = xfs_attr_leaf_getvalue(bp, args);
1211         xfs_da_brelse(args->trans, bp);
1212         if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1213                 error = xfs_attr_rmtval_get(args);
1214         }
1215         return(error);
1216 }
1217
1218 /*
1219  * Copy out attribute entries for attr_list(), for leaf attribute lists.
1220  */
1221 STATIC int
1222 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1223 {
1224         xfs_attr_leafblock_t *leaf;
1225         int error;
1226         xfs_dabuf_t *bp;
1227
1228         context->cursor->blkno = 0;
1229         error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1230         if (error)
1231                 return XFS_ERROR(error);
1232         ASSERT(bp != NULL);
1233         leaf = bp->data;
1234         if (unlikely(be16_to_cpu(leaf->hdr.info.magic) != XFS_ATTR_LEAF_MAGIC)) {
1235                 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1236                                      context->dp->i_mount, leaf);
1237                 xfs_da_brelse(NULL, bp);
1238                 return XFS_ERROR(EFSCORRUPTED);
1239         }
1240
1241         error = xfs_attr_leaf_list_int(bp, context);
1242         xfs_da_brelse(NULL, bp);
1243         return XFS_ERROR(error);
1244 }
1245
1246
1247 /*========================================================================
1248  * External routines when attribute list size > XFS_LBSIZE(mp).
1249  *========================================================================*/
1250
1251 /*
1252  * Add a name to a Btree-format attribute list.
1253  *
1254  * This will involve walking down the Btree, and may involve splitting
1255  * leaf nodes and even splitting intermediate nodes up to and including
1256  * the root node (a special case of an intermediate node).
1257  *
1258  * "Remote" attribute values confuse the issue and atomic rename operations
1259  * add a whole extra layer of confusion on top of that.
1260  */
1261 STATIC int
1262 xfs_attr_node_addname(xfs_da_args_t *args)
1263 {
1264         xfs_da_state_t *state;
1265         xfs_da_state_blk_t *blk;
1266         xfs_inode_t *dp;
1267         xfs_mount_t *mp;
1268         int committed, retval, error;
1269
1270         /*
1271          * Fill in bucket of arguments/results/context to carry around.
1272          */
1273         dp = args->dp;
1274         mp = dp->i_mount;
1275 restart:
1276         state = xfs_da_state_alloc();
1277         state->args = args;
1278         state->mp = mp;
1279         state->blocksize = state->mp->m_sb.sb_blocksize;
1280         state->node_ents = state->mp->m_attr_node_ents;
1281
1282         /*
1283          * Search to see if name already exists, and get back a pointer
1284          * to where it should go.
1285          */
1286         error = xfs_da_node_lookup_int(state, &retval);
1287         if (error)
1288                 goto out;
1289         blk = &state->path.blk[ state->path.active-1 ];
1290         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1291         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1292                 goto out;
1293         } else if (retval == EEXIST) {
1294                 if (args->flags & ATTR_CREATE)
1295                         goto out;
1296                 args->rename = 1;                       /* atomic rename op */
1297                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
1298                 args->index2 = args->index;
1299                 args->rmtblkno2 = args->rmtblkno;
1300                 args->rmtblkcnt2 = args->rmtblkcnt;
1301                 args->rmtblkno = 0;
1302                 args->rmtblkcnt = 0;
1303         }
1304
1305         retval = xfs_attr_leaf_add(blk->bp, state->args);
1306         if (retval == ENOSPC) {
1307                 if (state->path.active == 1) {
1308                         /*
1309                          * Its really a single leaf node, but it had
1310                          * out-of-line values so it looked like it *might*
1311                          * have been a b-tree.
1312                          */
1313                         xfs_da_state_free(state);
1314                         XFS_BMAP_INIT(args->flist, args->firstblock);
1315                         error = xfs_attr_leaf_to_node(args);
1316                         if (!error) {
1317                                 error = xfs_bmap_finish(&args->trans,
1318                                                         args->flist,
1319                                                         &committed);
1320                         }
1321                         if (error) {
1322                                 ASSERT(committed);
1323                                 args->trans = NULL;
1324                                 xfs_bmap_cancel(args->flist);
1325                                 goto out;
1326                         }
1327
1328                         /*
1329                          * bmap_finish() may have committed the last trans
1330                          * and started a new one.  We need the inode to be
1331                          * in all transactions.
1332                          */
1333                         if (committed) {
1334                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1335                                 xfs_trans_ihold(args->trans, dp);
1336                         }
1337
1338                         /*
1339                          * Commit the node conversion and start the next
1340                          * trans in the chain.
1341                          */
1342                         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1343                                 goto out;
1344
1345                         goto restart;
1346                 }
1347
1348                 /*
1349                  * Split as many Btree elements as required.
1350                  * This code tracks the new and old attr's location
1351                  * in the index/blkno/rmtblkno/rmtblkcnt fields and
1352                  * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1353                  */
1354                 XFS_BMAP_INIT(args->flist, args->firstblock);
1355                 error = xfs_da_split(state);
1356                 if (!error) {
1357                         error = xfs_bmap_finish(&args->trans, args->flist,
1358                                                 &committed);
1359                 }
1360                 if (error) {
1361                         ASSERT(committed);
1362                         args->trans = NULL;
1363                         xfs_bmap_cancel(args->flist);
1364                         goto out;
1365                 }
1366
1367                 /*
1368                  * bmap_finish() may have committed the last trans and started
1369                  * a new one.  We need the inode to be in all transactions.
1370                  */
1371                 if (committed) {
1372                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1373                         xfs_trans_ihold(args->trans, dp);
1374                 }
1375         } else {
1376                 /*
1377                  * Addition succeeded, update Btree hashvals.
1378                  */
1379                 xfs_da_fixhashpath(state, &state->path);
1380         }
1381
1382         /*
1383          * Kill the state structure, we're done with it and need to
1384          * allow the buffers to come back later.
1385          */
1386         xfs_da_state_free(state);
1387         state = NULL;
1388
1389         /*
1390          * Commit the leaf addition or btree split and start the next
1391          * trans in the chain.
1392          */
1393         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1394                 goto out;
1395
1396         /*
1397          * If there was an out-of-line value, allocate the blocks we
1398          * identified for its storage and copy the value.  This is done
1399          * after we create the attribute so that we don't overflow the
1400          * maximum size of a transaction and/or hit a deadlock.
1401          */
1402         if (args->rmtblkno > 0) {
1403                 error = xfs_attr_rmtval_set(args);
1404                 if (error)
1405                         return(error);
1406         }
1407
1408         /*
1409          * If this is an atomic rename operation, we must "flip" the
1410          * incomplete flags on the "new" and "old" attribute/value pairs
1411          * so that one disappears and one appears atomically.  Then we
1412          * must remove the "old" attribute/value pair.
1413          */
1414         if (args->rename) {
1415                 /*
1416                  * In a separate transaction, set the incomplete flag on the
1417                  * "old" attr and clear the incomplete flag on the "new" attr.
1418                  */
1419                 error = xfs_attr_leaf_flipflags(args);
1420                 if (error)
1421                         goto out;
1422
1423                 /*
1424                  * Dismantle the "old" attribute/value pair by removing
1425                  * a "remote" value (if it exists).
1426                  */
1427                 args->index = args->index2;
1428                 args->blkno = args->blkno2;
1429                 args->rmtblkno = args->rmtblkno2;
1430                 args->rmtblkcnt = args->rmtblkcnt2;
1431                 if (args->rmtblkno) {
1432                         error = xfs_attr_rmtval_remove(args);
1433                         if (error)
1434                                 return(error);
1435                 }
1436
1437                 /*
1438                  * Re-find the "old" attribute entry after any split ops.
1439                  * The INCOMPLETE flag means that we will find the "old"
1440                  * attr, not the "new" one.
1441                  */
1442                 args->flags |= XFS_ATTR_INCOMPLETE;
1443                 state = xfs_da_state_alloc();
1444                 state->args = args;
1445                 state->mp = mp;
1446                 state->blocksize = state->mp->m_sb.sb_blocksize;
1447                 state->node_ents = state->mp->m_attr_node_ents;
1448                 state->inleaf = 0;
1449                 error = xfs_da_node_lookup_int(state, &retval);
1450                 if (error)
1451                         goto out;
1452
1453                 /*
1454                  * Remove the name and update the hashvals in the tree.
1455                  */
1456                 blk = &state->path.blk[ state->path.active-1 ];
1457                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1458                 error = xfs_attr_leaf_remove(blk->bp, args);
1459                 xfs_da_fixhashpath(state, &state->path);
1460
1461                 /*
1462                  * Check to see if the tree needs to be collapsed.
1463                  */
1464                 if (retval && (state->path.active > 1)) {
1465                         XFS_BMAP_INIT(args->flist, args->firstblock);
1466                         error = xfs_da_join(state);
1467                         if (!error) {
1468                                 error = xfs_bmap_finish(&args->trans,
1469                                                         args->flist,
1470                                                         &committed);
1471                         }
1472                         if (error) {
1473                                 ASSERT(committed);
1474                                 args->trans = NULL;
1475                                 xfs_bmap_cancel(args->flist);
1476                                 goto out;
1477                         }
1478
1479                         /*
1480                          * bmap_finish() may have committed the last trans
1481                          * and started a new one.  We need the inode to be
1482                          * in all transactions.
1483                          */
1484                         if (committed) {
1485                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1486                                 xfs_trans_ihold(args->trans, dp);
1487                         }
1488                 }
1489
1490                 /*
1491                  * Commit and start the next trans in the chain.
1492                  */
1493                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1494                         goto out;
1495
1496         } else if (args->rmtblkno > 0) {
1497                 /*
1498                  * Added a "remote" value, just clear the incomplete flag.
1499                  */
1500                 error = xfs_attr_leaf_clearflag(args);
1501                 if (error)
1502                         goto out;
1503         }
1504         retval = error = 0;
1505
1506 out:
1507         if (state)
1508                 xfs_da_state_free(state);
1509         if (error)
1510                 return(error);
1511         return(retval);
1512 }
1513
1514 /*
1515  * Remove a name from a B-tree attribute list.
1516  *
1517  * This will involve walking down the Btree, and may involve joining
1518  * leaf nodes and even joining intermediate nodes up to and including
1519  * the root node (a special case of an intermediate node).
1520  */
1521 STATIC int
1522 xfs_attr_node_removename(xfs_da_args_t *args)
1523 {
1524         xfs_da_state_t *state;
1525         xfs_da_state_blk_t *blk;
1526         xfs_inode_t *dp;
1527         xfs_dabuf_t *bp;
1528         int retval, error, committed, forkoff;
1529
1530         /*
1531          * Tie a string around our finger to remind us where we are.
1532          */
1533         dp = args->dp;
1534         state = xfs_da_state_alloc();
1535         state->args = args;
1536         state->mp = dp->i_mount;
1537         state->blocksize = state->mp->m_sb.sb_blocksize;
1538         state->node_ents = state->mp->m_attr_node_ents;
1539
1540         /*
1541          * Search to see if name exists, and get back a pointer to it.
1542          */
1543         error = xfs_da_node_lookup_int(state, &retval);
1544         if (error || (retval != EEXIST)) {
1545                 if (error == 0)
1546                         error = retval;
1547                 goto out;
1548         }
1549
1550         /*
1551          * If there is an out-of-line value, de-allocate the blocks.
1552          * This is done before we remove the attribute so that we don't
1553          * overflow the maximum size of a transaction and/or hit a deadlock.
1554          */
1555         blk = &state->path.blk[ state->path.active-1 ];
1556         ASSERT(blk->bp != NULL);
1557         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1558         if (args->rmtblkno > 0) {
1559                 /*
1560                  * Fill in disk block numbers in the state structure
1561                  * so that we can get the buffers back after we commit
1562                  * several transactions in the following calls.
1563                  */
1564                 error = xfs_attr_fillstate(state);
1565                 if (error)
1566                         goto out;
1567
1568                 /*
1569                  * Mark the attribute as INCOMPLETE, then bunmapi() the
1570                  * remote value.
1571                  */
1572                 error = xfs_attr_leaf_setflag(args);
1573                 if (error)
1574                         goto out;
1575                 error = xfs_attr_rmtval_remove(args);
1576                 if (error)
1577                         goto out;
1578
1579                 /*
1580                  * Refill the state structure with buffers, the prior calls
1581                  * released our buffers.
1582                  */
1583                 error = xfs_attr_refillstate(state);
1584                 if (error)
1585                         goto out;
1586         }
1587
1588         /*
1589          * Remove the name and update the hashvals in the tree.
1590          */
1591         blk = &state->path.blk[ state->path.active-1 ];
1592         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1593         retval = xfs_attr_leaf_remove(blk->bp, args);
1594         xfs_da_fixhashpath(state, &state->path);
1595
1596         /*
1597          * Check to see if the tree needs to be collapsed.
1598          */
1599         if (retval && (state->path.active > 1)) {
1600                 XFS_BMAP_INIT(args->flist, args->firstblock);
1601                 error = xfs_da_join(state);
1602                 if (!error) {
1603                         error = xfs_bmap_finish(&args->trans, args->flist,
1604                                                 &committed);
1605                 }
1606                 if (error) {
1607                         ASSERT(committed);
1608                         args->trans = NULL;
1609                         xfs_bmap_cancel(args->flist);
1610                         goto out;
1611                 }
1612
1613                 /*
1614                  * bmap_finish() may have committed the last trans and started
1615                  * a new one.  We need the inode to be in all transactions.
1616                  */
1617                 if (committed) {
1618                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1619                         xfs_trans_ihold(args->trans, dp);
1620                 }
1621
1622                 /*
1623                  * Commit the Btree join operation and start a new trans.
1624                  */
1625                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1626                         goto out;
1627         }
1628
1629         /*
1630          * If the result is small enough, push it all into the inode.
1631          */
1632         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1633                 /*
1634                  * Have to get rid of the copy of this dabuf in the state.
1635                  */
1636                 ASSERT(state->path.active == 1);
1637                 ASSERT(state->path.blk[0].bp);
1638                 xfs_da_buf_done(state->path.blk[0].bp);
1639                 state->path.blk[0].bp = NULL;
1640
1641                 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1642                                                      XFS_ATTR_FORK);
1643                 if (error)
1644                         goto out;
1645                 ASSERT(be16_to_cpu(((xfs_attr_leafblock_t *)
1646                                       bp->data)->hdr.info.magic)
1647                                                        == XFS_ATTR_LEAF_MAGIC);
1648
1649                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1650                         XFS_BMAP_INIT(args->flist, args->firstblock);
1651                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1652                         /* bp is gone due to xfs_da_shrink_inode */
1653                         if (!error) {
1654                                 error = xfs_bmap_finish(&args->trans,
1655                                                         args->flist,
1656                                                         &committed);
1657                         }
1658                         if (error) {
1659                                 ASSERT(committed);
1660                                 args->trans = NULL;
1661                                 xfs_bmap_cancel(args->flist);
1662                                 goto out;
1663                         }
1664
1665                         /*
1666                          * bmap_finish() may have committed the last trans
1667                          * and started a new one.  We need the inode to be
1668                          * in all transactions.
1669                          */
1670                         if (committed) {
1671                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1672                                 xfs_trans_ihold(args->trans, dp);
1673                         }
1674                 } else
1675                         xfs_da_brelse(args->trans, bp);
1676         }
1677         error = 0;
1678
1679 out:
1680         xfs_da_state_free(state);
1681         return(error);
1682 }
1683
1684 /*
1685  * Fill in the disk block numbers in the state structure for the buffers
1686  * that are attached to the state structure.
1687  * This is done so that we can quickly reattach ourselves to those buffers
1688  * after some set of transaction commits have released these buffers.
1689  */
1690 STATIC int
1691 xfs_attr_fillstate(xfs_da_state_t *state)
1692 {
1693         xfs_da_state_path_t *path;
1694         xfs_da_state_blk_t *blk;
1695         int level;
1696
1697         /*
1698          * Roll down the "path" in the state structure, storing the on-disk
1699          * block number for those buffers in the "path".
1700          */
1701         path = &state->path;
1702         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1703         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1704                 if (blk->bp) {
1705                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1706                         xfs_da_buf_done(blk->bp);
1707                         blk->bp = NULL;
1708                 } else {
1709                         blk->disk_blkno = 0;
1710                 }
1711         }
1712
1713         /*
1714          * Roll down the "altpath" in the state structure, storing the on-disk
1715          * block number for those buffers in the "altpath".
1716          */
1717         path = &state->altpath;
1718         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1719         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1720                 if (blk->bp) {
1721                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1722                         xfs_da_buf_done(blk->bp);
1723                         blk->bp = NULL;
1724                 } else {
1725                         blk->disk_blkno = 0;
1726                 }
1727         }
1728
1729         return(0);
1730 }
1731
1732 /*
1733  * Reattach the buffers to the state structure based on the disk block
1734  * numbers stored in the state structure.
1735  * This is done after some set of transaction commits have released those
1736  * buffers from our grip.
1737  */
1738 STATIC int
1739 xfs_attr_refillstate(xfs_da_state_t *state)
1740 {
1741         xfs_da_state_path_t *path;
1742         xfs_da_state_blk_t *blk;
1743         int level, error;
1744
1745         /*
1746          * Roll down the "path" in the state structure, storing the on-disk
1747          * block number for those buffers in the "path".
1748          */
1749         path = &state->path;
1750         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1751         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1752                 if (blk->disk_blkno) {
1753                         error = xfs_da_read_buf(state->args->trans,
1754                                                 state->args->dp,
1755                                                 blk->blkno, blk->disk_blkno,
1756                                                 &blk->bp, XFS_ATTR_FORK);
1757                         if (error)
1758                                 return(error);
1759                 } else {
1760                         blk->bp = NULL;
1761                 }
1762         }
1763
1764         /*
1765          * Roll down the "altpath" in the state structure, storing the on-disk
1766          * block number for those buffers in the "altpath".
1767          */
1768         path = &state->altpath;
1769         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1770         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1771                 if (blk->disk_blkno) {
1772                         error = xfs_da_read_buf(state->args->trans,
1773                                                 state->args->dp,
1774                                                 blk->blkno, blk->disk_blkno,
1775                                                 &blk->bp, XFS_ATTR_FORK);
1776                         if (error)
1777                                 return(error);
1778                 } else {
1779                         blk->bp = NULL;
1780                 }
1781         }
1782
1783         return(0);
1784 }
1785
1786 /*
1787  * Look up a filename in a node attribute list.
1788  *
1789  * This routine gets called for any attribute fork that has more than one
1790  * block, ie: both true Btree attr lists and for single-leaf-blocks with
1791  * "remote" values taking up more blocks.
1792  */
1793 STATIC int
1794 xfs_attr_node_get(xfs_da_args_t *args)
1795 {
1796         xfs_da_state_t *state;
1797         xfs_da_state_blk_t *blk;
1798         int error, retval;
1799         int i;
1800
1801         state = xfs_da_state_alloc();
1802         state->args = args;
1803         state->mp = args->dp->i_mount;
1804         state->blocksize = state->mp->m_sb.sb_blocksize;
1805         state->node_ents = state->mp->m_attr_node_ents;
1806
1807         /*
1808          * Search to see if name exists, and get back a pointer to it.
1809          */
1810         error = xfs_da_node_lookup_int(state, &retval);
1811         if (error) {
1812                 retval = error;
1813         } else if (retval == EEXIST) {
1814                 blk = &state->path.blk[ state->path.active-1 ];
1815                 ASSERT(blk->bp != NULL);
1816                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1817
1818                 /*
1819                  * Get the value, local or "remote"
1820                  */
1821                 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1822                 if (!retval && (args->rmtblkno > 0)
1823                     && !(args->flags & ATTR_KERNOVAL)) {
1824                         retval = xfs_attr_rmtval_get(args);
1825                 }
1826         }
1827
1828         /*
1829          * If not in a transaction, we have to release all the buffers.
1830          */
1831         for (i = 0; i < state->path.active; i++) {
1832                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1833                 state->path.blk[i].bp = NULL;
1834         }
1835
1836         xfs_da_state_free(state);
1837         return(retval);
1838 }
1839
1840 STATIC int                                                      /* error */
1841 xfs_attr_node_list(xfs_attr_list_context_t *context)
1842 {
1843         attrlist_cursor_kern_t *cursor;
1844         xfs_attr_leafblock_t *leaf;
1845         xfs_da_intnode_t *node;
1846         xfs_da_node_entry_t *btree;
1847         int error, i;
1848         xfs_dabuf_t *bp;
1849
1850         cursor = context->cursor;
1851         cursor->initted = 1;
1852
1853         /*
1854          * Do all sorts of validation on the passed-in cursor structure.
1855          * If anything is amiss, ignore the cursor and look up the hashval
1856          * starting from the btree root.
1857          */
1858         bp = NULL;
1859         if (cursor->blkno > 0) {
1860                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1861                                               &bp, XFS_ATTR_FORK);
1862                 if ((error != 0) && (error != EFSCORRUPTED))
1863                         return(error);
1864                 if (bp) {
1865                         node = bp->data;
1866                         switch (be16_to_cpu(node->hdr.info.magic)) {
1867                         case XFS_DA_NODE_MAGIC:
1868                                 xfs_attr_trace_l_cn("wrong blk", context, node);
1869                                 xfs_da_brelse(NULL, bp);
1870                                 bp = NULL;
1871                                 break;
1872                         case XFS_ATTR_LEAF_MAGIC:
1873                                 leaf = bp->data;
1874                                 if (cursor->hashval > be32_to_cpu(leaf->entries[
1875                                     be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1876                                         xfs_attr_trace_l_cl("wrong blk",
1877                                                            context, leaf);
1878                                         xfs_da_brelse(NULL, bp);
1879                                         bp = NULL;
1880                                 } else if (cursor->hashval <=
1881                                              be32_to_cpu(leaf->entries[0].hashval)) {
1882                                         xfs_attr_trace_l_cl("maybe wrong blk",
1883                                                            context, leaf);
1884                                         xfs_da_brelse(NULL, bp);
1885                                         bp = NULL;
1886                                 }
1887                                 break;
1888                         default:
1889                                 xfs_attr_trace_l_c("wrong blk - ??", context);
1890                                 xfs_da_brelse(NULL, bp);
1891                                 bp = NULL;
1892                         }
1893                 }
1894         }
1895
1896         /*
1897          * We did not find what we expected given the cursor's contents,
1898          * so we start from the top and work down based on the hash value.
1899          * Note that start of node block is same as start of leaf block.
1900          */
1901         if (bp == NULL) {
1902                 cursor->blkno = 0;
1903                 for (;;) {
1904                         error = xfs_da_read_buf(NULL, context->dp,
1905                                                       cursor->blkno, -1, &bp,
1906                                                       XFS_ATTR_FORK);
1907                         if (error)
1908                                 return(error);
1909                         if (unlikely(bp == NULL)) {
1910                                 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1911                                                  XFS_ERRLEVEL_LOW,
1912                                                  context->dp->i_mount);
1913                                 return(XFS_ERROR(EFSCORRUPTED));
1914                         }
1915                         node = bp->data;
1916                         if (be16_to_cpu(node->hdr.info.magic)
1917                                                         == XFS_ATTR_LEAF_MAGIC)
1918                                 break;
1919                         if (unlikely(be16_to_cpu(node->hdr.info.magic)
1920                                                         != XFS_DA_NODE_MAGIC)) {
1921                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1922                                                      XFS_ERRLEVEL_LOW,
1923                                                      context->dp->i_mount,
1924                                                      node);
1925                                 xfs_da_brelse(NULL, bp);
1926                                 return(XFS_ERROR(EFSCORRUPTED));
1927                         }
1928                         btree = node->btree;
1929                         for (i = 0; i < be16_to_cpu(node->hdr.count);
1930                                                                 btree++, i++) {
1931                                 if (cursor->hashval
1932                                                 <= be32_to_cpu(btree->hashval)) {
1933                                         cursor->blkno = be32_to_cpu(btree->before);
1934                                         xfs_attr_trace_l_cb("descending",
1935                                                             context, btree);
1936                                         break;
1937                                 }
1938                         }
1939                         if (i == be16_to_cpu(node->hdr.count)) {
1940                                 xfs_da_brelse(NULL, bp);
1941                                 return(0);
1942                         }
1943                         xfs_da_brelse(NULL, bp);
1944                 }
1945         }
1946         ASSERT(bp != NULL);
1947
1948         /*
1949          * Roll upward through the blocks, processing each leaf block in
1950          * order.  As long as there is space in the result buffer, keep
1951          * adding the information.
1952          */
1953         for (;;) {
1954                 leaf = bp->data;
1955                 if (unlikely(be16_to_cpu(leaf->hdr.info.magic)
1956                                                 != XFS_ATTR_LEAF_MAGIC)) {
1957                         XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1958                                              XFS_ERRLEVEL_LOW,
1959                                              context->dp->i_mount, leaf);
1960                         xfs_da_brelse(NULL, bp);
1961                         return(XFS_ERROR(EFSCORRUPTED));
1962                 }
1963                 error = xfs_attr_leaf_list_int(bp, context);
1964                 if (error) {
1965                         xfs_da_brelse(NULL, bp);
1966                         return error;
1967                 }
1968                 if (context->seen_enough || leaf->hdr.info.forw == 0)
1969                         break;
1970                 cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1971                 xfs_da_brelse(NULL, bp);
1972                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1973                                               &bp, XFS_ATTR_FORK);
1974                 if (error)
1975                         return(error);
1976                 if (unlikely((bp == NULL))) {
1977                         XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1978                                          XFS_ERRLEVEL_LOW,
1979                                          context->dp->i_mount);
1980                         return(XFS_ERROR(EFSCORRUPTED));
1981                 }
1982         }
1983         xfs_da_brelse(NULL, bp);
1984         return(0);
1985 }
1986
1987
1988 /*========================================================================
1989  * External routines for manipulating out-of-line attribute values.
1990  *========================================================================*/
1991
1992 /*
1993  * Read the value associated with an attribute from the out-of-line buffer
1994  * that we stored it in.
1995  */
1996 int
1997 xfs_attr_rmtval_get(xfs_da_args_t *args)
1998 {
1999         xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
2000         xfs_mount_t *mp;
2001         xfs_daddr_t dblkno;
2002         xfs_caddr_t dst;
2003         xfs_buf_t *bp;
2004         int nmap, error, tmp, valuelen, blkcnt, i;
2005         xfs_dablk_t lblkno;
2006
2007         ASSERT(!(args->flags & ATTR_KERNOVAL));
2008
2009         mp = args->dp->i_mount;
2010         dst = args->value;
2011         valuelen = args->valuelen;
2012         lblkno = args->rmtblkno;
2013         while (valuelen > 0) {
2014                 nmap = ATTR_RMTVALUE_MAPSIZE;
2015                 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
2016                                   args->rmtblkcnt,
2017                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2018                                   NULL, 0, map, &nmap, NULL, NULL);
2019                 if (error)
2020                         return(error);
2021                 ASSERT(nmap >= 1);
2022
2023                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
2024                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
2025                                (map[i].br_startblock != HOLESTARTBLOCK));
2026                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
2027                         blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
2028                         error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
2029                                              blkcnt, XFS_BUF_LOCK, &bp);
2030                         if (error)
2031                                 return(error);
2032
2033                         tmp = (valuelen < XFS_BUF_SIZE(bp))
2034                                 ? valuelen : XFS_BUF_SIZE(bp);
2035                         xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
2036                         xfs_buf_relse(bp);
2037                         dst += tmp;
2038                         valuelen -= tmp;
2039
2040                         lblkno += map[i].br_blockcount;
2041                 }
2042         }
2043         ASSERT(valuelen == 0);
2044         return(0);
2045 }
2046
2047 /*
2048  * Write the value associated with an attribute into the out-of-line buffer
2049  * that we have defined for it.
2050  */
2051 STATIC int
2052 xfs_attr_rmtval_set(xfs_da_args_t *args)
2053 {
2054         xfs_mount_t *mp;
2055         xfs_fileoff_t lfileoff;
2056         xfs_inode_t *dp;
2057         xfs_bmbt_irec_t map;
2058         xfs_daddr_t dblkno;
2059         xfs_caddr_t src;
2060         xfs_buf_t *bp;
2061         xfs_dablk_t lblkno;
2062         int blkcnt, valuelen, nmap, error, tmp, committed;
2063
2064         dp = args->dp;
2065         mp = dp->i_mount;
2066         src = args->value;
2067
2068         /*
2069          * Find a "hole" in the attribute address space large enough for
2070          * us to drop the new attribute's value into.
2071          */
2072         blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2073         lfileoff = 0;
2074         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2075                                                    XFS_ATTR_FORK);
2076         if (error) {
2077                 return(error);
2078         }
2079         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2080         args->rmtblkcnt = blkcnt;
2081
2082         /*
2083          * Roll through the "value", allocating blocks on disk as required.
2084          */
2085         while (blkcnt > 0) {
2086                 /*
2087                  * Allocate a single extent, up to the size of the value.
2088                  */
2089                 XFS_BMAP_INIT(args->flist, args->firstblock);
2090                 nmap = 1;
2091                 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2092                                   blkcnt,
2093                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2094                                                         XFS_BMAPI_WRITE,
2095                                   args->firstblock, args->total, &map, &nmap,
2096                                   args->flist, NULL);
2097                 if (!error) {
2098                         error = xfs_bmap_finish(&args->trans, args->flist,
2099                                                 &committed);
2100                 }
2101                 if (error) {
2102                         ASSERT(committed);
2103                         args->trans = NULL;
2104                         xfs_bmap_cancel(args->flist);
2105                         return(error);
2106                 }
2107
2108                 /*
2109                  * bmap_finish() may have committed the last trans and started
2110                  * a new one.  We need the inode to be in all transactions.
2111                  */
2112                 if (committed) {
2113                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2114                         xfs_trans_ihold(args->trans, dp);
2115                 }
2116
2117                 ASSERT(nmap == 1);
2118                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2119                        (map.br_startblock != HOLESTARTBLOCK));
2120                 lblkno += map.br_blockcount;
2121                 blkcnt -= map.br_blockcount;
2122
2123                 /*
2124                  * Start the next trans in the chain.
2125                  */
2126                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
2127                         return (error);
2128         }
2129
2130         /*
2131          * Roll through the "value", copying the attribute value to the
2132          * already-allocated blocks.  Blocks are written synchronously
2133          * so that we can know they are all on disk before we turn off
2134          * the INCOMPLETE flag.
2135          */
2136         lblkno = args->rmtblkno;
2137         valuelen = args->valuelen;
2138         while (valuelen > 0) {
2139                 /*
2140                  * Try to remember where we decided to put the value.
2141                  */
2142                 XFS_BMAP_INIT(args->flist, args->firstblock);
2143                 nmap = 1;
2144                 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2145                                   args->rmtblkcnt,
2146                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2147                                   args->firstblock, 0, &map, &nmap,
2148                                   NULL, NULL);
2149                 if (error) {
2150                         return(error);
2151                 }
2152                 ASSERT(nmap == 1);
2153                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2154                        (map.br_startblock != HOLESTARTBLOCK));
2155
2156                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2157                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2158
2159                 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno,
2160                                                         blkcnt, XFS_BUF_LOCK);
2161                 ASSERT(bp);
2162                 ASSERT(!XFS_BUF_GETERROR(bp));
2163
2164                 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2165                                                         XFS_BUF_SIZE(bp);
2166                 xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2167                 if (tmp < XFS_BUF_SIZE(bp))
2168                         xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2169                 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2170                         return (error);
2171                 }
2172                 src += tmp;
2173                 valuelen -= tmp;
2174
2175                 lblkno += map.br_blockcount;
2176         }
2177         ASSERT(valuelen == 0);
2178         return(0);
2179 }
2180
2181 /*
2182  * Remove the value associated with an attribute by deleting the
2183  * out-of-line buffer that it is stored on.
2184  */
2185 STATIC int
2186 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2187 {
2188         xfs_mount_t *mp;
2189         xfs_bmbt_irec_t map;
2190         xfs_buf_t *bp;
2191         xfs_daddr_t dblkno;
2192         xfs_dablk_t lblkno;
2193         int valuelen, blkcnt, nmap, error, done, committed;
2194
2195         mp = args->dp->i_mount;
2196
2197         /*
2198          * Roll through the "value", invalidating the attribute value's
2199          * blocks.
2200          */
2201         lblkno = args->rmtblkno;
2202         valuelen = args->rmtblkcnt;
2203         while (valuelen > 0) {
2204                 /*
2205                  * Try to remember where we decided to put the value.
2206                  */
2207                 XFS_BMAP_INIT(args->flist, args->firstblock);
2208                 nmap = 1;
2209                 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2210                                         args->rmtblkcnt,
2211                                         XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2212                                         args->firstblock, 0, &map, &nmap,
2213                                         args->flist, NULL);
2214                 if (error) {
2215                         return(error);
2216                 }
2217                 ASSERT(nmap == 1);
2218                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2219                        (map.br_startblock != HOLESTARTBLOCK));
2220
2221                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2222                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2223
2224                 /*
2225                  * If the "remote" value is in the cache, remove it.
2226                  */
2227                 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt,
2228                                 XFS_INCORE_TRYLOCK);
2229                 if (bp) {
2230                         XFS_BUF_STALE(bp);
2231                         XFS_BUF_UNDELAYWRITE(bp);
2232                         xfs_buf_relse(bp);
2233                         bp = NULL;
2234                 }
2235
2236                 valuelen -= map.br_blockcount;
2237
2238                 lblkno += map.br_blockcount;
2239         }
2240
2241         /*
2242          * Keep de-allocating extents until the remote-value region is gone.
2243          */
2244         lblkno = args->rmtblkno;
2245         blkcnt = args->rmtblkcnt;
2246         done = 0;
2247         while (!done) {
2248                 XFS_BMAP_INIT(args->flist, args->firstblock);
2249                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2250                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2251                                     1, args->firstblock, args->flist,
2252                                     NULL, &done);
2253                 if (!error) {
2254                         error = xfs_bmap_finish(&args->trans, args->flist,
2255                                                 &committed);
2256                 }
2257                 if (error) {
2258                         ASSERT(committed);
2259                         args->trans = NULL;
2260                         xfs_bmap_cancel(args->flist);
2261                         return(error);
2262                 }
2263
2264                 /*
2265                  * bmap_finish() may have committed the last trans and started
2266                  * a new one.  We need the inode to be in all transactions.
2267                  */
2268                 if (committed) {
2269                         xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2270                         xfs_trans_ihold(args->trans, args->dp);
2271                 }
2272
2273                 /*
2274                  * Close out trans and start the next one in the chain.
2275                  */
2276                 if ((error = xfs_attr_rolltrans(&args->trans, args->dp)))
2277                         return (error);
2278         }
2279         return(0);
2280 }
2281
2282 #if defined(XFS_ATTR_TRACE)
2283 /*
2284  * Add a trace buffer entry for an attr_list context structure.
2285  */
2286 void
2287 xfs_attr_trace_l_c(char *where, struct xfs_attr_list_context *context)
2288 {
2289         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_C, where,
2290                 (__psunsigned_t)context->dp,
2291                 (__psunsigned_t)context->cursor->hashval,
2292                 (__psunsigned_t)context->cursor->blkno,
2293                 (__psunsigned_t)context->cursor->offset,
2294                 (__psunsigned_t)context->alist,
2295                 (__psunsigned_t)context->bufsize,
2296                 (__psunsigned_t)context->count,
2297                 (__psunsigned_t)context->firstu,
2298                 (__psunsigned_t)
2299                         ((context->count > 0) &&
2300                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2301                                 ? (ATTR_ENTRY(context->alist,
2302                                               context->count-1)->a_valuelen)
2303                                 : 0,
2304                 (__psunsigned_t)context->dupcnt,
2305                 (__psunsigned_t)context->flags,
2306                 (__psunsigned_t)NULL,
2307                 (__psunsigned_t)NULL,
2308                 (__psunsigned_t)NULL);
2309 }
2310
2311 /*
2312  * Add a trace buffer entry for a context structure and a Btree node.
2313  */
2314 void
2315 xfs_attr_trace_l_cn(char *where, struct xfs_attr_list_context *context,
2316                          struct xfs_da_intnode *node)
2317 {
2318         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CN, where,
2319                 (__psunsigned_t)context->dp,
2320                 (__psunsigned_t)context->cursor->hashval,
2321                 (__psunsigned_t)context->cursor->blkno,
2322                 (__psunsigned_t)context->cursor->offset,
2323                 (__psunsigned_t)context->alist,
2324                 (__psunsigned_t)context->bufsize,
2325                 (__psunsigned_t)context->count,
2326                 (__psunsigned_t)context->firstu,
2327                 (__psunsigned_t)
2328                         ((context->count > 0) &&
2329                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2330                                 ? (ATTR_ENTRY(context->alist,
2331                                               context->count-1)->a_valuelen)
2332                                 : 0,
2333                 (__psunsigned_t)context->dupcnt,
2334                 (__psunsigned_t)context->flags,
2335                 (__psunsigned_t)be16_to_cpu(node->hdr.count),
2336                 (__psunsigned_t)be32_to_cpu(node->btree[0].hashval),
2337                 (__psunsigned_t)be32_to_cpu(node->btree[
2338                                     be16_to_cpu(node->hdr.count)-1].hashval));
2339 }
2340
2341 /*
2342  * Add a trace buffer entry for a context structure and a Btree element.
2343  */
2344 void
2345 xfs_attr_trace_l_cb(char *where, struct xfs_attr_list_context *context,
2346                           struct xfs_da_node_entry *btree)
2347 {
2348         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CB, where,
2349                 (__psunsigned_t)context->dp,
2350                 (__psunsigned_t)context->cursor->hashval,
2351                 (__psunsigned_t)context->cursor->blkno,
2352                 (__psunsigned_t)context->cursor->offset,
2353                 (__psunsigned_t)context->alist,
2354                 (__psunsigned_t)context->bufsize,
2355                 (__psunsigned_t)context->count,
2356                 (__psunsigned_t)context->firstu,
2357                 (__psunsigned_t)
2358                         ((context->count > 0) &&
2359                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2360                                 ? (ATTR_ENTRY(context->alist,
2361                                               context->count-1)->a_valuelen)
2362                                 : 0,
2363                 (__psunsigned_t)context->dupcnt,
2364                 (__psunsigned_t)context->flags,
2365                 (__psunsigned_t)be32_to_cpu(btree->hashval),
2366                 (__psunsigned_t)be32_to_cpu(btree->before),
2367                 (__psunsigned_t)NULL);
2368 }
2369
2370 /*
2371  * Add a trace buffer entry for a context structure and a leaf block.
2372  */
2373 void
2374 xfs_attr_trace_l_cl(char *where, struct xfs_attr_list_context *context,
2375                               struct xfs_attr_leafblock *leaf)
2376 {
2377         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CL, where,
2378                 (__psunsigned_t)context->dp,
2379                 (__psunsigned_t)context->cursor->hashval,
2380                 (__psunsigned_t)context->cursor->blkno,
2381                 (__psunsigned_t)context->cursor->offset,
2382                 (__psunsigned_t)context->alist,
2383                 (__psunsigned_t)context->bufsize,
2384                 (__psunsigned_t)context->count,
2385                 (__psunsigned_t)context->firstu,
2386                 (__psunsigned_t)
2387                         ((context->count > 0) &&
2388                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2389                                 ? (ATTR_ENTRY(context->alist,
2390                                               context->count-1)->a_valuelen)
2391                                 : 0,
2392                 (__psunsigned_t)context->dupcnt,
2393                 (__psunsigned_t)context->flags,
2394                 (__psunsigned_t)be16_to_cpu(leaf->hdr.count),
2395                 (__psunsigned_t)be32_to_cpu(leaf->entries[0].hashval),
2396                 (__psunsigned_t)be32_to_cpu(leaf->entries[
2397                                 be16_to_cpu(leaf->hdr.count)-1].hashval));
2398 }
2399
2400 /*
2401  * Add a trace buffer entry for the arguments given to the routine,
2402  * generic form.
2403  */
2404 void
2405 xfs_attr_trace_enter(int type, char *where,
2406                          __psunsigned_t a2, __psunsigned_t a3,
2407                          __psunsigned_t a4, __psunsigned_t a5,
2408                          __psunsigned_t a6, __psunsigned_t a7,
2409                          __psunsigned_t a8, __psunsigned_t a9,
2410                          __psunsigned_t a10, __psunsigned_t a11,
2411                          __psunsigned_t a12, __psunsigned_t a13,
2412                          __psunsigned_t a14, __psunsigned_t a15)
2413 {
2414         ASSERT(xfs_attr_trace_buf);
2415         ktrace_enter(xfs_attr_trace_buf, (void *)((__psunsigned_t)type),
2416                                          (void *)where,
2417                                          (void *)a2,  (void *)a3,  (void *)a4,
2418                                          (void *)a5,  (void *)a6,  (void *)a7,
2419                                          (void *)a8,  (void *)a9,  (void *)a10,
2420                                          (void *)a11, (void *)a12, (void *)a13,
2421                                          (void *)a14, (void *)a15);
2422 }
2423 #endif  /* XFS_ATTR_TRACE */
2424
2425
2426 /*========================================================================
2427  * System (pseudo) namespace attribute interface routines.
2428  *========================================================================*/
2429
2430 STATIC int
2431 posix_acl_access_set(
2432         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2433 {
2434         return xfs_acl_vset(vp, data, size, _ACL_TYPE_ACCESS);
2435 }
2436
2437 STATIC int
2438 posix_acl_access_remove(
2439         bhv_vnode_t *vp, char *name, int xflags)
2440 {
2441         return xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
2442 }
2443
2444 STATIC int
2445 posix_acl_access_get(
2446         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2447 {
2448         return xfs_acl_vget(vp, data, size, _ACL_TYPE_ACCESS);
2449 }
2450
2451 STATIC int
2452 posix_acl_access_exists(
2453         bhv_vnode_t *vp)
2454 {
2455         return xfs_acl_vhasacl_access(vp);
2456 }
2457
2458 STATIC int
2459 posix_acl_default_set(
2460         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2461 {
2462         return xfs_acl_vset(vp, data, size, _ACL_TYPE_DEFAULT);
2463 }
2464
2465 STATIC int
2466 posix_acl_default_get(
2467         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2468 {
2469         return xfs_acl_vget(vp, data, size, _ACL_TYPE_DEFAULT);
2470 }
2471
2472 STATIC int
2473 posix_acl_default_remove(
2474         bhv_vnode_t *vp, char *name, int xflags)
2475 {
2476         return xfs_acl_vremove(vp, _ACL_TYPE_DEFAULT);
2477 }
2478
2479 STATIC int
2480 posix_acl_default_exists(
2481         bhv_vnode_t *vp)
2482 {
2483         return xfs_acl_vhasacl_default(vp);
2484 }
2485
2486 static struct attrnames posix_acl_access = {
2487         .attr_name      = "posix_acl_access",
2488         .attr_namelen   = sizeof("posix_acl_access") - 1,
2489         .attr_get       = posix_acl_access_get,
2490         .attr_set       = posix_acl_access_set,
2491         .attr_remove    = posix_acl_access_remove,
2492         .attr_exists    = posix_acl_access_exists,
2493 };
2494
2495 static struct attrnames posix_acl_default = {
2496         .attr_name      = "posix_acl_default",
2497         .attr_namelen   = sizeof("posix_acl_default") - 1,
2498         .attr_get       = posix_acl_default_get,
2499         .attr_set       = posix_acl_default_set,
2500         .attr_remove    = posix_acl_default_remove,
2501         .attr_exists    = posix_acl_default_exists,
2502 };
2503
2504 static struct attrnames *attr_system_names[] =
2505         { &posix_acl_access, &posix_acl_default };
2506
2507
2508 /*========================================================================
2509  * Namespace-prefix-style attribute name interface routines.
2510  *========================================================================*/
2511
2512 STATIC int
2513 attr_generic_set(
2514         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2515 {
2516         return -xfs_attr_set(xfs_vtoi(vp), name, data, size, xflags);
2517 }
2518
2519 STATIC int
2520 attr_generic_get(
2521         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2522 {
2523         int     error, asize = size;
2524
2525         error = xfs_attr_get(xfs_vtoi(vp), name, data,
2526                                     &asize, xflags, NULL);
2527         if (!error)
2528                 return asize;
2529         return -error;
2530 }
2531
2532 STATIC int
2533 attr_generic_remove(
2534         bhv_vnode_t *vp, char *name, int xflags)
2535 {
2536         return -xfs_attr_remove(xfs_vtoi(vp), name, xflags);
2537 }
2538
2539 STATIC int
2540 attr_generic_listadd(
2541         attrnames_t             *prefix,
2542         attrnames_t             *namesp,
2543         void                    *data,
2544         size_t                  size,
2545         ssize_t                 *result)
2546 {
2547         char                    *p = data + *result;
2548
2549         *result += prefix->attr_namelen;
2550         *result += namesp->attr_namelen + 1;
2551         if (!size)
2552                 return 0;
2553         if (*result > size)
2554                 return -ERANGE;
2555         strcpy(p, prefix->attr_name);
2556         p += prefix->attr_namelen;
2557         strcpy(p, namesp->attr_name);
2558         p += namesp->attr_namelen + 1;
2559         return 0;
2560 }
2561
2562 STATIC int
2563 attr_system_list(
2564         bhv_vnode_t             *vp,
2565         void                    *data,
2566         size_t                  size,
2567         ssize_t                 *result)
2568 {
2569         attrnames_t             *namesp;
2570         int                     i, error = 0;
2571
2572         for (i = 0; i < ATTR_SYSCOUNT; i++) {
2573                 namesp = attr_system_names[i];
2574                 if (!namesp->attr_exists || !namesp->attr_exists(vp))
2575                         continue;
2576                 error = attr_generic_listadd(&attr_system, namesp,
2577                                                 data, size, result);
2578                 if (error)
2579                         break;
2580         }
2581         return error;
2582 }
2583
2584 int
2585 attr_generic_list(
2586         bhv_vnode_t *vp, void *data, size_t size, int xflags, ssize_t *result)
2587 {
2588         attrlist_cursor_kern_t  cursor = { 0 };
2589         int                     error;
2590
2591         error = xfs_attr_list(xfs_vtoi(vp), data, size, xflags, &cursor);
2592         if (error > 0)
2593                 return -error;
2594         *result = -error;
2595         return attr_system_list(vp, data, size, result);
2596 }
2597
2598 attrnames_t *
2599 attr_lookup_namespace(
2600         char                    *name,
2601         struct attrnames        **names,
2602         int                     nnames)
2603 {
2604         int                     i;
2605
2606         for (i = 0; i < nnames; i++)
2607                 if (!strncmp(name, names[i]->attr_name, names[i]->attr_namelen))
2608                         return names[i];
2609         return NULL;
2610 }
2611
2612 /*
2613  * Some checks to prevent people abusing EAs to get over quota:
2614  * - Don't allow modifying user EAs on devices/symlinks;
2615  * - Don't allow modifying user EAs if sticky bit set;
2616  */
2617 STATIC int
2618 attr_user_capable(
2619         bhv_vnode_t     *vp,
2620         cred_t          *cred)
2621 {
2622         struct inode    *inode = vn_to_inode(vp);
2623
2624         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2625                 return -EPERM;
2626         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
2627             !capable(CAP_SYS_ADMIN))
2628                 return -EPERM;
2629         if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
2630             (current_fsuid(cred) != inode->i_uid) && !capable(CAP_FOWNER))
2631                 return -EPERM;
2632         return 0;
2633 }
2634
2635 STATIC int
2636 attr_trusted_capable(
2637         bhv_vnode_t     *vp,
2638         cred_t          *cred)
2639 {
2640         struct inode    *inode = vn_to_inode(vp);
2641
2642         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2643                 return -EPERM;
2644         if (!capable(CAP_SYS_ADMIN))
2645                 return -EPERM;
2646         return 0;
2647 }
2648
2649 STATIC int
2650 attr_secure_capable(
2651         bhv_vnode_t     *vp,
2652         cred_t          *cred)
2653 {
2654         return -ENOSECURITY;
2655 }
2656
2657 STATIC int
2658 attr_system_set(
2659         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2660 {
2661         attrnames_t     *namesp;
2662         int             error;
2663
2664         if (xflags & ATTR_CREATE)
2665                 return -EINVAL;
2666
2667         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2668         if (!namesp)
2669                 return -EOPNOTSUPP;
2670         error = namesp->attr_set(vp, name, data, size, xflags);
2671         if (!error)
2672                 error = vn_revalidate(vp);
2673         return error;
2674 }
2675
2676 STATIC int
2677 attr_system_get(
2678         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2679 {
2680         attrnames_t     *namesp;
2681
2682         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2683         if (!namesp)
2684                 return -EOPNOTSUPP;
2685         return namesp->attr_get(vp, name, data, size, xflags);
2686 }
2687
2688 STATIC int
2689 attr_system_remove(
2690         bhv_vnode_t *vp, char *name, int xflags)
2691 {
2692         attrnames_t     *namesp;
2693
2694         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2695         if (!namesp)
2696                 return -EOPNOTSUPP;
2697         return namesp->attr_remove(vp, name, xflags);
2698 }
2699
2700 struct attrnames attr_system = {
2701         .attr_name      = "system.",
2702         .attr_namelen   = sizeof("system.") - 1,
2703         .attr_flag      = ATTR_SYSTEM,
2704         .attr_get       = attr_system_get,
2705         .attr_set       = attr_system_set,
2706         .attr_remove    = attr_system_remove,
2707         .attr_capable   = (attrcapable_t)fs_noerr,
2708 };
2709
2710 struct attrnames attr_trusted = {
2711         .attr_name      = "trusted.",
2712         .attr_namelen   = sizeof("trusted.") - 1,
2713         .attr_flag      = ATTR_ROOT,
2714         .attr_get       = attr_generic_get,
2715         .attr_set       = attr_generic_set,
2716         .attr_remove    = attr_generic_remove,
2717         .attr_capable   = attr_trusted_capable,
2718 };
2719
2720 struct attrnames attr_secure = {
2721         .attr_name      = "security.",
2722         .attr_namelen   = sizeof("security.") - 1,
2723         .attr_flag      = ATTR_SECURE,
2724         .attr_get       = attr_generic_get,
2725         .attr_set       = attr_generic_set,
2726         .attr_remove    = attr_generic_remove,
2727         .attr_capable   = attr_secure_capable,
2728 };
2729
2730 struct attrnames attr_user = {
2731         .attr_name      = "user.",
2732         .attr_namelen   = sizeof("user.") - 1,
2733         .attr_get       = attr_generic_get,
2734         .attr_set       = attr_generic_set,
2735         .attr_remove    = attr_generic_remove,
2736         .attr_capable   = attr_user_capable,
2737 };
2738
2739 struct attrnames *attr_namespaces[] =
2740         { &attr_system, &attr_trusted, &attr_secure, &attr_user };