[XFS] Ondisk format extension for extended attributes (attr2). Basically,
[linux-2.6] / fs / xfs / xfs_dir.c
1 /*
2  * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32
33 #include "xfs.h"
34
35 #include "xfs_macros.h"
36 #include "xfs_types.h"
37 #include "xfs_inum.h"
38 #include "xfs_log.h"
39 #include "xfs_trans.h"
40 #include "xfs_sb.h"
41 #include "xfs_dir.h"
42 #include "xfs_dir2.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_alloc.h"
49 #include "xfs_btree.h"
50 #include "xfs_attr_sf.h"
51 #include "xfs_dir_sf.h"
52 #include "xfs_dir2_sf.h"
53 #include "xfs_dinode.h"
54 #include "xfs_inode.h"
55 #include "xfs_bmap.h"
56 #include "xfs_da_btree.h"
57 #include "xfs_dir_leaf.h"
58 #include "xfs_error.h"
59
60 /*
61  * xfs_dir.c
62  *
63  * Provide the external interfaces to manage directories.
64  */
65
66 /*========================================================================
67  * Function prototypes for the kernel.
68  *========================================================================*/
69
70 /*
71  * Functions for the dirops interfaces.
72  */
73 static void     xfs_dir_mount(struct xfs_mount *mp);
74
75 static int      xfs_dir_isempty(struct xfs_inode *dp);
76
77 static int      xfs_dir_init(struct xfs_trans *trans,
78                              struct xfs_inode *dir,
79                              struct xfs_inode *parent_dir);
80
81 static int      xfs_dir_createname(struct xfs_trans *trans,
82                                    struct xfs_inode *dp,
83                                    char *name_string,
84                                    int name_len,
85                                    xfs_ino_t inode_number,
86                                    xfs_fsblock_t *firstblock,
87                                    xfs_bmap_free_t *flist,
88                                    xfs_extlen_t total);
89
90 static int      xfs_dir_lookup(struct xfs_trans *tp,
91                                struct xfs_inode *dp,
92                                char *name_string,
93                                int name_length,
94                                xfs_ino_t *inode_number);
95
96 static int      xfs_dir_removename(struct xfs_trans *trans,
97                                    struct xfs_inode *dp,
98                                    char *name_string,
99                                    int name_length,
100                                    xfs_ino_t ino,
101                                    xfs_fsblock_t *firstblock,
102                                    xfs_bmap_free_t *flist,
103                                    xfs_extlen_t total);
104
105 static int      xfs_dir_getdents(struct xfs_trans *tp,
106                                  struct xfs_inode *dp,
107                                  struct uio *uiop,
108                                  int *eofp);
109
110 static int      xfs_dir_replace(struct xfs_trans *tp,
111                                 struct xfs_inode *dp,
112                                 char *name_string,
113                                 int name_length,
114                                 xfs_ino_t inode_number,
115                                 xfs_fsblock_t *firstblock,
116                                 xfs_bmap_free_t *flist,
117                                 xfs_extlen_t total);
118
119 static int      xfs_dir_canenter(struct xfs_trans *tp,
120                                  struct xfs_inode *dp,
121                                  char *name_string,
122                                  int name_length);
123
124 static int      xfs_dir_shortform_validate_ondisk(xfs_mount_t *mp,
125                                                   xfs_dinode_t *dip);
126
127 xfs_dirops_t xfsv1_dirops = {
128         .xd_mount                       = xfs_dir_mount,
129         .xd_isempty                     = xfs_dir_isempty,
130         .xd_init                        = xfs_dir_init,
131         .xd_createname                  = xfs_dir_createname,
132         .xd_lookup                      = xfs_dir_lookup,
133         .xd_removename                  = xfs_dir_removename,
134         .xd_getdents                    = xfs_dir_getdents,
135         .xd_replace                     = xfs_dir_replace,
136         .xd_canenter                    = xfs_dir_canenter,
137         .xd_shortform_validate_ondisk   = xfs_dir_shortform_validate_ondisk,
138         .xd_shortform_to_single         = xfs_dir_shortform_to_leaf,
139 };
140
141 /*
142  * Internal routines when dirsize == XFS_LBSIZE(mp).
143  */
144 STATIC int xfs_dir_leaf_lookup(xfs_da_args_t *args);
145 STATIC int xfs_dir_leaf_removename(xfs_da_args_t *args, int *number_entries,
146                                                  int *total_namebytes);
147 STATIC int xfs_dir_leaf_getdents(xfs_trans_t *trans, xfs_inode_t *dp,
148                                              uio_t *uio, int *eofp,
149                                              xfs_dirent_t *dbp,
150                                              xfs_dir_put_t put);
151 STATIC int xfs_dir_leaf_replace(xfs_da_args_t *args);
152
153 /*
154  * Internal routines when dirsize > XFS_LBSIZE(mp).
155  */
156 STATIC int xfs_dir_node_addname(xfs_da_args_t *args);
157 STATIC int xfs_dir_node_lookup(xfs_da_args_t *args);
158 STATIC int xfs_dir_node_removename(xfs_da_args_t *args);
159 STATIC int xfs_dir_node_getdents(xfs_trans_t *trans, xfs_inode_t *dp,
160                                              uio_t *uio, int *eofp,
161                                              xfs_dirent_t *dbp,
162                                              xfs_dir_put_t put);
163 STATIC int xfs_dir_node_replace(xfs_da_args_t *args);
164
165 #if defined(XFS_DIR_TRACE)
166 ktrace_t *xfs_dir_trace_buf;
167 #endif
168
169
170 /*========================================================================
171  * Overall external interface routines.
172  *========================================================================*/
173
174 xfs_dahash_t    xfs_dir_hash_dot, xfs_dir_hash_dotdot;
175
176 /*
177  * One-time startup routine called from xfs_init().
178  */
179 void
180 xfs_dir_startup(void)
181 {
182         xfs_dir_hash_dot = xfs_da_hashname(".", 1);
183         xfs_dir_hash_dotdot = xfs_da_hashname("..", 2);
184 }
185
186 /*
187  * Initialize directory-related fields in the mount structure.
188  */
189 static void
190 xfs_dir_mount(xfs_mount_t *mp)
191 {
192         uint shortcount, leafcount, count;
193
194         mp->m_dirversion = 1;
195         if (mp->m_flags & XFS_MOUNT_COMPAT_ATTR) {
196                 shortcount = (mp->m_attroffset -
197                                 (uint)sizeof(xfs_dir_sf_hdr_t)) /
198                                  (uint)sizeof(xfs_dir_sf_entry_t);
199                 leafcount = (XFS_LBSIZE(mp) -
200                                 (uint)sizeof(xfs_dir_leaf_hdr_t)) /
201                                  ((uint)sizeof(xfs_dir_leaf_entry_t) +
202                                   (uint)sizeof(xfs_dir_leaf_name_t));
203         } else {
204                 shortcount = (XFS_BMDR_SPACE_CALC(MINABTPTRS) -
205                               (uint)sizeof(xfs_dir_sf_hdr_t)) /
206                                (uint)sizeof(xfs_dir_sf_entry_t);
207                 leafcount = (XFS_LBSIZE(mp) -
208                             (uint)sizeof(xfs_dir_leaf_hdr_t)) /
209                              ((uint)sizeof(xfs_dir_leaf_entry_t) +
210                               (uint)sizeof(xfs_dir_leaf_name_t));
211         }
212         count = shortcount > leafcount ? shortcount : leafcount;
213         mp->m_dircook_elog = xfs_da_log2_roundup(count + 1);
214         ASSERT(mp->m_dircook_elog <= mp->m_sb.sb_blocklog);
215         mp->m_dir_node_ents = mp->m_attr_node_ents =
216                 (XFS_LBSIZE(mp) - (uint)sizeof(xfs_da_node_hdr_t)) /
217                 (uint)sizeof(xfs_da_node_entry_t);
218         mp->m_dir_magicpct = (XFS_LBSIZE(mp) * 37) / 100;
219         mp->m_dirblksize = mp->m_sb.sb_blocksize;
220         mp->m_dirblkfsbs = 1;
221 }
222
223 /*
224  * Return 1 if directory contains only "." and "..".
225  */
226 static int
227 xfs_dir_isempty(xfs_inode_t *dp)
228 {
229         xfs_dir_sf_hdr_t *hdr;
230
231         ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
232         if (dp->i_d.di_size == 0)
233                 return(1);
234         if (dp->i_d.di_size > XFS_IFORK_DSIZE(dp))
235                 return(0);
236         hdr = (xfs_dir_sf_hdr_t *)dp->i_df.if_u1.if_data;
237         return(hdr->count == 0);
238 }
239
240 /*
241  * Initialize a directory with its "." and ".." entries.
242  */
243 static int
244 xfs_dir_init(xfs_trans_t *trans, xfs_inode_t *dir, xfs_inode_t *parent_dir)
245 {
246         xfs_da_args_t args;
247         int error;
248
249         memset((char *)&args, 0, sizeof(args));
250         args.dp = dir;
251         args.trans = trans;
252
253         ASSERT((dir->i_d.di_mode & S_IFMT) == S_IFDIR);
254         if ((error = xfs_dir_ino_validate(trans->t_mountp, parent_dir->i_ino)))
255                 return error;
256
257         return(xfs_dir_shortform_create(&args, parent_dir->i_ino));
258 }
259
260 /*
261  * Generic handler routine to add a name to a directory.
262  * Transitions directory from shortform to Btree as necessary.
263  */
264 static int                                                      /* error */
265 xfs_dir_createname(xfs_trans_t *trans, xfs_inode_t *dp, char *name,
266                    int namelen, xfs_ino_t inum, xfs_fsblock_t *firstblock,
267                    xfs_bmap_free_t *flist, xfs_extlen_t total)
268 {
269         xfs_da_args_t args;
270         int retval, newsize, done;
271
272         ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
273
274         if ((retval = xfs_dir_ino_validate(trans->t_mountp, inum)))
275                 return (retval);
276
277         XFS_STATS_INC(xs_dir_create);
278         /*
279          * Fill in the arg structure for this request.
280          */
281         args.name = name;
282         args.namelen = namelen;
283         args.hashval = xfs_da_hashname(name, namelen);
284         args.inumber = inum;
285         args.dp = dp;
286         args.firstblock = firstblock;
287         args.flist = flist;
288         args.total = total;
289         args.whichfork = XFS_DATA_FORK;
290         args.trans = trans;
291         args.justcheck = 0;
292         args.addname = args.oknoent = 1;
293
294         /*
295          * Decide on what work routines to call based on the inode size.
296          */
297         done = 0;
298         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
299                 newsize = XFS_DIR_SF_ENTSIZE_BYNAME(args.namelen);
300                 if ((dp->i_d.di_size + newsize) <= XFS_IFORK_DSIZE(dp)) {
301                         retval = xfs_dir_shortform_addname(&args);
302                         done = 1;
303                 } else {
304                         if (total == 0)
305                                 return XFS_ERROR(ENOSPC);
306                         retval = xfs_dir_shortform_to_leaf(&args);
307                         done = retval != 0;
308                 }
309         }
310         if (!done && xfs_bmap_one_block(dp, XFS_DATA_FORK)) {
311                 retval = xfs_dir_leaf_addname(&args);
312                 done = retval != ENOSPC;
313                 if (!done) {
314                         if (total == 0)
315                                 return XFS_ERROR(ENOSPC);
316                         retval = xfs_dir_leaf_to_node(&args);
317                         done = retval != 0;
318                 }
319         }
320         if (!done) {
321                 retval = xfs_dir_node_addname(&args);
322         }
323         return(retval);
324 }
325
326 /*
327  * Generic handler routine to check if a name can be added to a directory,
328  * without adding any blocks to the directory.
329  */
330 static int                                                      /* error */
331 xfs_dir_canenter(xfs_trans_t *trans, xfs_inode_t *dp, char *name, int namelen)
332 {
333         xfs_da_args_t args;
334         int retval, newsize;
335
336         ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
337         /*
338          * Fill in the arg structure for this request.
339          */
340         args.name = name;
341         args.namelen = namelen;
342         args.hashval = xfs_da_hashname(name, namelen);
343         args.inumber = 0;
344         args.dp = dp;
345         args.firstblock = NULL;
346         args.flist = NULL;
347         args.total = 0;
348         args.whichfork = XFS_DATA_FORK;
349         args.trans = trans;
350         args.justcheck = args.addname = args.oknoent = 1;
351
352         /*
353          * Decide on what work routines to call based on the inode size.
354          */
355         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
356                 newsize = XFS_DIR_SF_ENTSIZE_BYNAME(args.namelen);
357                 if ((dp->i_d.di_size + newsize) <= XFS_IFORK_DSIZE(dp))
358                         retval = 0;
359                 else
360                         retval = XFS_ERROR(ENOSPC);
361         } else if (xfs_bmap_one_block(dp, XFS_DATA_FORK)) {
362                 retval = xfs_dir_leaf_addname(&args);
363         } else {
364                 retval = xfs_dir_node_addname(&args);
365         }
366         return(retval);
367 }
368
369 /*
370  * Generic handler routine to remove a name from a directory.
371  * Transitions directory from Btree to shortform as necessary.
372  */
373 static int                                                      /* error */
374 xfs_dir_removename(xfs_trans_t *trans, xfs_inode_t *dp, char *name,
375                    int namelen, xfs_ino_t ino, xfs_fsblock_t *firstblock,
376                    xfs_bmap_free_t *flist, xfs_extlen_t total)
377 {
378         xfs_da_args_t args;
379         int count, totallen, newsize, retval;
380
381         ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
382         XFS_STATS_INC(xs_dir_remove);
383         /*
384          * Fill in the arg structure for this request.
385          */
386         args.name = name;
387         args.namelen = namelen;
388         args.hashval = xfs_da_hashname(name, namelen);
389         args.inumber = ino;
390         args.dp = dp;
391         args.firstblock = firstblock;
392         args.flist = flist;
393         args.total = total;
394         args.whichfork = XFS_DATA_FORK;
395         args.trans = trans;
396         args.justcheck = args.addname = args.oknoent = 0;
397
398         /*
399          * Decide on what work routines to call based on the inode size.
400          */
401         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
402                 retval = xfs_dir_shortform_removename(&args);
403         } else if (xfs_bmap_one_block(dp, XFS_DATA_FORK)) {
404                 retval = xfs_dir_leaf_removename(&args, &count, &totallen);
405                 if (retval == 0) {
406                         newsize = XFS_DIR_SF_ALLFIT(count, totallen);
407                         if (newsize <= XFS_IFORK_DSIZE(dp)) {
408                                 retval = xfs_dir_leaf_to_shortform(&args);
409                         }
410                 }
411         } else {
412                 retval = xfs_dir_node_removename(&args);
413         }
414         return(retval);
415 }
416
417 static int                                                      /* error */
418 xfs_dir_lookup(xfs_trans_t *trans, xfs_inode_t *dp, char *name, int namelen,
419                                    xfs_ino_t *inum)
420 {
421         xfs_da_args_t args;
422         int retval;
423
424         ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
425
426         XFS_STATS_INC(xs_dir_lookup);
427         /*
428          * Fill in the arg structure for this request.
429          */
430         args.name = name;
431         args.namelen = namelen;
432         args.hashval = xfs_da_hashname(name, namelen);
433         args.inumber = 0;
434         args.dp = dp;
435         args.firstblock = NULL;
436         args.flist = NULL;
437         args.total = 0;
438         args.whichfork = XFS_DATA_FORK;
439         args.trans = trans;
440         args.justcheck = args.addname = 0;
441         args.oknoent = 1;
442
443         /*
444          * Decide on what work routines to call based on the inode size.
445          */
446         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
447                 retval = xfs_dir_shortform_lookup(&args);
448         } else if (xfs_bmap_one_block(dp, XFS_DATA_FORK)) {
449                 retval = xfs_dir_leaf_lookup(&args);
450         } else {
451                 retval = xfs_dir_node_lookup(&args);
452         }
453         if (retval == EEXIST)
454                 retval = 0;
455         *inum = args.inumber;
456         return(retval);
457 }
458
459 /*
460  * Implement readdir.
461  */
462 static int                                                      /* error */
463 xfs_dir_getdents(xfs_trans_t *trans, xfs_inode_t *dp, uio_t *uio, int *eofp)
464 {
465         xfs_dirent_t *dbp;
466         int  alignment, retval;
467         xfs_dir_put_t put;
468
469         XFS_STATS_INC(xs_dir_getdents);
470         ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
471
472         /*
473          * If our caller has given us a single contiguous memory buffer,
474          * just work directly within that buffer.  If it's in user memory,
475          * lock it down first.
476          */
477         alignment = sizeof(xfs_off_t) - 1;
478         if ((uio->uio_iovcnt == 1) &&
479             (((__psint_t)uio->uio_iov[0].iov_base & alignment) == 0) &&
480             ((uio->uio_iov[0].iov_len & alignment) == 0)) {
481                 dbp = NULL;
482                 put = xfs_dir_put_dirent64_direct;
483         } else {
484                 dbp = kmem_alloc(sizeof(*dbp) + MAXNAMELEN, KM_SLEEP);
485                 put = xfs_dir_put_dirent64_uio;
486         }
487
488         /*
489          * Decide on what work routines to call based on the inode size.
490          */
491         *eofp = 0;
492
493         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
494                 retval = xfs_dir_shortform_getdents(dp, uio, eofp, dbp, put);
495         } else if (xfs_bmap_one_block(dp, XFS_DATA_FORK)) {
496                 retval = xfs_dir_leaf_getdents(trans, dp, uio, eofp, dbp, put);
497         } else {
498                 retval = xfs_dir_node_getdents(trans, dp, uio, eofp, dbp, put);
499         }
500         if (dbp != NULL)
501                 kmem_free(dbp, sizeof(*dbp) + MAXNAMELEN);
502
503         return(retval);
504 }
505
506 static int                                                      /* error */
507 xfs_dir_replace(xfs_trans_t *trans, xfs_inode_t *dp, char *name, int namelen,
508                                     xfs_ino_t inum, xfs_fsblock_t *firstblock,
509                                     xfs_bmap_free_t *flist, xfs_extlen_t total)
510 {
511         xfs_da_args_t args;
512         int retval;
513
514         ASSERT((dp->i_d.di_mode & S_IFMT) == S_IFDIR);
515
516         if ((retval = xfs_dir_ino_validate(trans->t_mountp, inum)))
517                 return retval;
518
519         /*
520          * Fill in the arg structure for this request.
521          */
522         args.name = name;
523         args.namelen = namelen;
524         args.hashval = xfs_da_hashname(name, namelen);
525         args.inumber = inum;
526         args.dp = dp;
527         args.firstblock = firstblock;
528         args.flist = flist;
529         args.total = total;
530         args.whichfork = XFS_DATA_FORK;
531         args.trans = trans;
532         args.justcheck = args.addname = args.oknoent = 0;
533
534         /*
535          * Decide on what work routines to call based on the inode size.
536          */
537         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
538                 retval = xfs_dir_shortform_replace(&args);
539         } else if (xfs_bmap_one_block(dp, XFS_DATA_FORK)) {
540                 retval = xfs_dir_leaf_replace(&args);
541         } else {
542                 retval = xfs_dir_node_replace(&args);
543         }
544
545         return(retval);
546 }
547
548 static int
549 xfs_dir_shortform_validate_ondisk(xfs_mount_t *mp, xfs_dinode_t *dp)
550 {
551         xfs_ino_t               ino;
552         int                     namelen_sum;
553         int                     count;
554         xfs_dir_shortform_t     *sf;
555         xfs_dir_sf_entry_t      *sfe;
556         int                     i;
557
558
559
560         if ((INT_GET(dp->di_core.di_mode, ARCH_CONVERT) & S_IFMT) != S_IFDIR) {
561                 return 0;
562         }
563         if (INT_GET(dp->di_core.di_format, ARCH_CONVERT) != XFS_DINODE_FMT_LOCAL) {
564                 return 0;
565         }
566         if (INT_GET(dp->di_core.di_size, ARCH_CONVERT) < sizeof(sf->hdr)) {
567                 xfs_fs_cmn_err(CE_WARN, mp, "Invalid shortform size: dp 0x%p",
568                         dp);
569                 return 1;
570         }
571         sf = (xfs_dir_shortform_t *)(&dp->di_u.di_dirsf);
572         ino = XFS_GET_DIR_INO8(sf->hdr.parent);
573         if (xfs_dir_ino_validate(mp, ino))
574                 return 1;
575
576         count = sf->hdr.count;
577         if ((count < 0) || ((count * 10) > XFS_LITINO(mp))) {
578                 xfs_fs_cmn_err(CE_WARN, mp,
579                         "Invalid shortform count: dp 0x%p", dp);
580                 return(1);
581         }
582
583         if (count == 0) {
584                 return 0;
585         }
586
587         namelen_sum = 0;
588         sfe = &sf->list[0];
589         for (i = sf->hdr.count - 1; i >= 0; i--) {
590                 ino = XFS_GET_DIR_INO8(sfe->inumber);
591                 xfs_dir_ino_validate(mp, ino);
592                 if (sfe->namelen >= XFS_LITINO(mp)) {
593                         xfs_fs_cmn_err(CE_WARN, mp,
594                                 "Invalid shortform namelen: dp 0x%p", dp);
595                         return 1;
596                 }
597                 namelen_sum += sfe->namelen;
598                 sfe = XFS_DIR_SF_NEXTENTRY(sfe);
599         }
600         if (namelen_sum >= XFS_LITINO(mp)) {
601                 xfs_fs_cmn_err(CE_WARN, mp,
602                         "Invalid shortform namelen: dp 0x%p", dp);
603                 return 1;
604         }
605
606         return 0;
607 }
608
609 /*========================================================================
610  * External routines when dirsize == XFS_LBSIZE(dp->i_mount).
611  *========================================================================*/
612
613 /*
614  * Add a name to the leaf directory structure
615  * This is the external routine.
616  */
617 int
618 xfs_dir_leaf_addname(xfs_da_args_t *args)
619 {
620         int index, retval;
621         xfs_dabuf_t *bp;
622
623         retval = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
624                                               XFS_DATA_FORK);
625         if (retval)
626                 return(retval);
627         ASSERT(bp != NULL);
628
629         retval = xfs_dir_leaf_lookup_int(bp, args, &index);
630         if (retval == ENOENT)
631                 retval = xfs_dir_leaf_add(bp, args, index);
632         xfs_da_buf_done(bp);
633         return(retval);
634 }
635
636 /*
637  * Remove a name from the leaf directory structure
638  * This is the external routine.
639  */
640 STATIC int
641 xfs_dir_leaf_removename(xfs_da_args_t *args, int *count, int *totallen)
642 {
643         xfs_dir_leafblock_t *leaf;
644         int index, retval;
645         xfs_dabuf_t *bp;
646
647         retval = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
648                                               XFS_DATA_FORK);
649         if (retval)
650                 return(retval);
651         ASSERT(bp != NULL);
652         leaf = bp->data;
653         ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR_LEAF_MAGIC);
654         retval = xfs_dir_leaf_lookup_int(bp, args, &index);
655         if (retval == EEXIST) {
656                 (void)xfs_dir_leaf_remove(args->trans, bp, index);
657                 *count = INT_GET(leaf->hdr.count, ARCH_CONVERT);
658                 *totallen = INT_GET(leaf->hdr.namebytes, ARCH_CONVERT);
659                 retval = 0;
660         }
661         xfs_da_buf_done(bp);
662         return(retval);
663 }
664
665 /*
666  * Look up a name in a leaf directory structure.
667  * This is the external routine.
668  */
669 STATIC int
670 xfs_dir_leaf_lookup(xfs_da_args_t *args)
671 {
672         int index, retval;
673         xfs_dabuf_t *bp;
674
675         retval = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
676                                               XFS_DATA_FORK);
677         if (retval)
678                 return(retval);
679         ASSERT(bp != NULL);
680         retval = xfs_dir_leaf_lookup_int(bp, args, &index);
681         xfs_da_brelse(args->trans, bp);
682         return(retval);
683 }
684
685 /*
686  * Copy out directory entries for getdents(), for leaf directories.
687  */
688 STATIC int
689 xfs_dir_leaf_getdents(xfs_trans_t *trans, xfs_inode_t *dp, uio_t *uio,
690                                   int *eofp, xfs_dirent_t *dbp, xfs_dir_put_t put)
691 {
692         xfs_dabuf_t *bp;
693         int retval, eob;
694
695         retval = xfs_da_read_buf(dp->i_transp, dp, 0, -1, &bp, XFS_DATA_FORK);
696         if (retval)
697                 return(retval);
698         ASSERT(bp != NULL);
699         retval = xfs_dir_leaf_getdents_int(bp, dp, 0, uio, &eob, dbp, put, -1);
700         xfs_da_brelse(trans, bp);
701         *eofp = (eob == 0);
702         return(retval);
703 }
704
705 /*
706  * Look up a name in a leaf directory structure, replace the inode number.
707  * This is the external routine.
708  */
709 STATIC int
710 xfs_dir_leaf_replace(xfs_da_args_t *args)
711 {
712         int index, retval;
713         xfs_dabuf_t *bp;
714         xfs_ino_t inum;
715         xfs_dir_leafblock_t *leaf;
716         xfs_dir_leaf_entry_t *entry;
717         xfs_dir_leaf_name_t *namest;
718
719         inum = args->inumber;
720         retval = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
721                                               XFS_DATA_FORK);
722         if (retval)
723                 return(retval);
724         ASSERT(bp != NULL);
725         retval = xfs_dir_leaf_lookup_int(bp, args, &index);
726         if (retval == EEXIST) {
727                 leaf = bp->data;
728                 entry = &leaf->entries[index];
729                 namest = XFS_DIR_LEAF_NAMESTRUCT(leaf, INT_GET(entry->nameidx, ARCH_CONVERT));
730                 /* XXX - replace assert? */
731                 XFS_DIR_SF_PUT_DIRINO(&inum, &namest->inumber);
732                 xfs_da_log_buf(args->trans, bp,
733                     XFS_DA_LOGRANGE(leaf, namest, sizeof(namest->inumber)));
734                 xfs_da_buf_done(bp);
735                 retval = 0;
736         } else
737                 xfs_da_brelse(args->trans, bp);
738         return(retval);
739 }
740
741
742 /*========================================================================
743  * External routines when dirsize > XFS_LBSIZE(mp).
744  *========================================================================*/
745
746 /*
747  * Add a name to a Btree-format directory.
748  *
749  * This will involve walking down the Btree, and may involve splitting
750  * leaf nodes and even splitting intermediate nodes up to and including
751  * the root node (a special case of an intermediate node).
752  */
753 STATIC int
754 xfs_dir_node_addname(xfs_da_args_t *args)
755 {
756         xfs_da_state_t *state;
757         xfs_da_state_blk_t *blk;
758         int retval, error;
759
760         /*
761          * Fill in bucket of arguments/results/context to carry around.
762          */
763         state = xfs_da_state_alloc();
764         state->args = args;
765         state->mp = args->dp->i_mount;
766         state->blocksize = state->mp->m_sb.sb_blocksize;
767         state->node_ents = state->mp->m_dir_node_ents;
768
769         /*
770          * Search to see if name already exists, and get back a pointer
771          * to where it should go.
772          */
773         error = xfs_da_node_lookup_int(state, &retval);
774         if (error)
775                 retval = error;
776         if (retval != ENOENT)
777                 goto error;
778         blk = &state->path.blk[ state->path.active-1 ];
779         ASSERT(blk->magic == XFS_DIR_LEAF_MAGIC);
780         retval = xfs_dir_leaf_add(blk->bp, args, blk->index);
781         if (retval == 0) {
782                 /*
783                  * Addition succeeded, update Btree hashvals.
784                  */
785                 if (!args->justcheck)
786                         xfs_da_fixhashpath(state, &state->path);
787         } else {
788                 /*
789                  * Addition failed, split as many Btree elements as required.
790                  */
791                 if (args->total == 0) {
792                         ASSERT(retval == ENOSPC);
793                         goto error;
794                 }
795                 retval = xfs_da_split(state);
796         }
797 error:
798         xfs_da_state_free(state);
799
800         return(retval);
801 }
802
803 /*
804  * Remove a name from a B-tree directory.
805  *
806  * This will involve walking down the Btree, and may involve joining
807  * leaf nodes and even joining intermediate nodes up to and including
808  * the root node (a special case of an intermediate node).
809  */
810 STATIC int
811 xfs_dir_node_removename(xfs_da_args_t *args)
812 {
813         xfs_da_state_t *state;
814         xfs_da_state_blk_t *blk;
815         int retval, error;
816
817         state = xfs_da_state_alloc();
818         state->args = args;
819         state->mp = args->dp->i_mount;
820         state->blocksize = state->mp->m_sb.sb_blocksize;
821         state->node_ents = state->mp->m_dir_node_ents;
822
823         /*
824          * Search to see if name exists, and get back a pointer to it.
825          */
826         error = xfs_da_node_lookup_int(state, &retval);
827         if (error)
828                 retval = error;
829         if (retval != EEXIST) {
830                 xfs_da_state_free(state);
831                 return(retval);
832         }
833
834         /*
835          * Remove the name and update the hashvals in the tree.
836          */
837         blk = &state->path.blk[ state->path.active-1 ];
838         ASSERT(blk->magic == XFS_DIR_LEAF_MAGIC);
839         retval = xfs_dir_leaf_remove(args->trans, blk->bp, blk->index);
840         xfs_da_fixhashpath(state, &state->path);
841
842         /*
843          * Check to see if the tree needs to be collapsed.
844          */
845         error = 0;
846         if (retval) {
847                 error = xfs_da_join(state);
848         }
849
850         xfs_da_state_free(state);
851         if (error)
852                 return(error);
853         return(0);
854 }
855
856 /*
857  * Look up a filename in a int directory.
858  * Use an internal routine to actually do all the work.
859  */
860 STATIC int
861 xfs_dir_node_lookup(xfs_da_args_t *args)
862 {
863         xfs_da_state_t *state;
864         int retval, error, i;
865
866         state = xfs_da_state_alloc();
867         state->args = args;
868         state->mp = args->dp->i_mount;
869         state->blocksize = state->mp->m_sb.sb_blocksize;
870         state->node_ents = state->mp->m_dir_node_ents;
871
872         /*
873          * Search to see if name exists,
874          * and get back a pointer to it.
875          */
876         error = xfs_da_node_lookup_int(state, &retval);
877         if (error) {
878                 retval = error;
879         }
880
881         /*
882          * If not in a transaction, we have to release all the buffers.
883          */
884         for (i = 0; i < state->path.active; i++) {
885                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
886                 state->path.blk[i].bp = NULL;
887         }
888
889         xfs_da_state_free(state);
890         return(retval);
891 }
892
893 STATIC int
894 xfs_dir_node_getdents(xfs_trans_t *trans, xfs_inode_t *dp, uio_t *uio,
895                                   int *eofp, xfs_dirent_t *dbp, xfs_dir_put_t put)
896 {
897         xfs_da_intnode_t *node;
898         xfs_da_node_entry_t *btree;
899         xfs_dir_leafblock_t *leaf = NULL;
900         xfs_dablk_t bno, nextbno;
901         xfs_dahash_t cookhash;
902         xfs_mount_t *mp;
903         int error, eob, i;
904         xfs_dabuf_t *bp;
905         xfs_daddr_t nextda;
906
907         /*
908          * Pick up our context.
909          */
910         mp = dp->i_mount;
911         bp = NULL;
912         bno = XFS_DA_COOKIE_BNO(mp, uio->uio_offset);
913         cookhash = XFS_DA_COOKIE_HASH(mp, uio->uio_offset);
914
915         xfs_dir_trace_g_du("node: start", dp, uio);
916
917         /*
918          * Re-find our place, even if we're confused about what our place is.
919          *
920          * First we check the block number from the magic cookie, it is a
921          * cache of where we ended last time.  If we find a leaf block, and
922          * the starting hashval in that block is less than our desired
923          * hashval, then we run with it.
924          */
925         if (bno > 0) {
926                 error = xfs_da_read_buf(trans, dp, bno, -2, &bp, XFS_DATA_FORK);
927                 if ((error != 0) && (error != EFSCORRUPTED))
928                         return(error);
929                 if (bp)
930                         leaf = bp->data;
931                 if (bp && INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) != XFS_DIR_LEAF_MAGIC) {
932                         xfs_dir_trace_g_dub("node: block not a leaf",
933                                                    dp, uio, bno);
934                         xfs_da_brelse(trans, bp);
935                         bp = NULL;
936                 }
937                 if (bp && INT_GET(leaf->entries[0].hashval, ARCH_CONVERT) > cookhash) {
938                         xfs_dir_trace_g_dub("node: leaf hash too large",
939                                                    dp, uio, bno);
940                         xfs_da_brelse(trans, bp);
941                         bp = NULL;
942                 }
943                 if (bp &&
944                     cookhash > INT_GET(leaf->entries[INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1].hashval, ARCH_CONVERT)) {
945                         xfs_dir_trace_g_dub("node: leaf hash too small",
946                                                    dp, uio, bno);
947                         xfs_da_brelse(trans, bp);
948                         bp = NULL;
949                 }
950         }
951
952         /*
953          * If we did not find a leaf block from the blockno in the cookie,
954          * or we there was no blockno in the cookie (eg: first time thru),
955          * the we start at the top of the Btree and re-find our hashval.
956          */
957         if (bp == NULL) {
958                 xfs_dir_trace_g_du("node: start at root" , dp, uio);
959                 bno = 0;
960                 for (;;) {
961                         error = xfs_da_read_buf(trans, dp, bno, -1, &bp,
962                                                        XFS_DATA_FORK);
963                         if (error)
964                                 return(error);
965                         if (bp == NULL)
966                                 return(XFS_ERROR(EFSCORRUPTED));
967                         node = bp->data;
968                         if (INT_GET(node->hdr.info.magic, ARCH_CONVERT) != XFS_DA_NODE_MAGIC)
969                                 break;
970                         btree = &node->btree[0];
971                         xfs_dir_trace_g_dun("node: node detail", dp, uio, node);
972                         for (i = 0; i < INT_GET(node->hdr.count, ARCH_CONVERT); btree++, i++) {
973                                 if (INT_GET(btree->hashval, ARCH_CONVERT) >= cookhash) {
974                                         bno = INT_GET(btree->before, ARCH_CONVERT);
975                                         break;
976                                 }
977                         }
978                         if (i == INT_GET(node->hdr.count, ARCH_CONVERT)) {
979                                 xfs_da_brelse(trans, bp);
980                                 xfs_dir_trace_g_du("node: hash beyond EOF",
981                                                           dp, uio);
982                                 uio->uio_offset = XFS_DA_MAKE_COOKIE(mp, 0, 0,
983                                                              XFS_DA_MAXHASH);
984                                 *eofp = 1;
985                                 return(0);
986                         }
987                         xfs_dir_trace_g_dub("node: going to block",
988                                                    dp, uio, bno);
989                         xfs_da_brelse(trans, bp);
990                 }
991         }
992         ASSERT(cookhash != XFS_DA_MAXHASH);
993
994         /*
995          * We've dropped down to the (first) leaf block that contains the
996          * hashval we are interested in.  Continue rolling upward thru the
997          * leaf blocks until we fill up our buffer.
998          */
999         for (;;) {
1000                 leaf = bp->data;
1001                 if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) != XFS_DIR_LEAF_MAGIC)) {
1002                         xfs_dir_trace_g_dul("node: not a leaf", dp, uio, leaf);
1003                         xfs_da_brelse(trans, bp);
1004                         XFS_CORRUPTION_ERROR("xfs_dir_node_getdents(1)",
1005                                              XFS_ERRLEVEL_LOW, mp, leaf);
1006                         return XFS_ERROR(EFSCORRUPTED);
1007                 }
1008                 xfs_dir_trace_g_dul("node: leaf detail", dp, uio, leaf);
1009                 if ((nextbno = INT_GET(leaf->hdr.info.forw, ARCH_CONVERT))) {
1010                         nextda = xfs_da_reada_buf(trans, dp, nextbno,
1011                                                   XFS_DATA_FORK);
1012                 } else
1013                         nextda = -1;
1014                 error = xfs_dir_leaf_getdents_int(bp, dp, bno, uio, &eob, dbp,
1015                                                   put, nextda);
1016                 xfs_da_brelse(trans, bp);
1017                 bno = nextbno;
1018                 if (eob) {
1019                         xfs_dir_trace_g_dub("node: E-O-B", dp, uio, bno);
1020                         *eofp = 0;
1021                         return(error);
1022                 }
1023                 if (bno == 0)
1024                         break;
1025                 error = xfs_da_read_buf(trans, dp, bno, nextda, &bp,
1026                                         XFS_DATA_FORK);
1027                 if (error)
1028                         return(error);
1029                 if (unlikely(bp == NULL)) {
1030                         XFS_ERROR_REPORT("xfs_dir_node_getdents(2)",
1031                                          XFS_ERRLEVEL_LOW, mp);
1032                         return(XFS_ERROR(EFSCORRUPTED));
1033                 }
1034         }
1035         *eofp = 1;
1036         xfs_dir_trace_g_du("node: E-O-F", dp, uio);
1037         return(0);
1038 }
1039
1040 /*
1041  * Look up a filename in an int directory, replace the inode number.
1042  * Use an internal routine to actually do the lookup.
1043  */
1044 STATIC int
1045 xfs_dir_node_replace(xfs_da_args_t *args)
1046 {
1047         xfs_da_state_t *state;
1048         xfs_da_state_blk_t *blk;
1049         xfs_dir_leafblock_t *leaf;
1050         xfs_dir_leaf_entry_t *entry;
1051         xfs_dir_leaf_name_t *namest;
1052         xfs_ino_t inum;
1053         int retval, error, i;
1054         xfs_dabuf_t *bp;
1055
1056         state = xfs_da_state_alloc();
1057         state->args = args;
1058         state->mp = args->dp->i_mount;
1059         state->blocksize = state->mp->m_sb.sb_blocksize;
1060         state->node_ents = state->mp->m_dir_node_ents;
1061         inum = args->inumber;
1062
1063         /*
1064          * Search to see if name exists,
1065          * and get back a pointer to it.
1066          */
1067         error = xfs_da_node_lookup_int(state, &retval);
1068         if (error) {
1069                 retval = error;
1070         }
1071
1072         if (retval == EEXIST) {
1073                 blk = &state->path.blk[state->path.active - 1];
1074                 ASSERT(blk->magic == XFS_DIR_LEAF_MAGIC);
1075                 bp = blk->bp;
1076                 leaf = bp->data;
1077                 entry = &leaf->entries[blk->index];
1078                 namest = XFS_DIR_LEAF_NAMESTRUCT(leaf, INT_GET(entry->nameidx, ARCH_CONVERT));
1079                 /* XXX - replace assert ? */
1080                 XFS_DIR_SF_PUT_DIRINO(&inum, &namest->inumber);
1081                 xfs_da_log_buf(args->trans, bp,
1082                     XFS_DA_LOGRANGE(leaf, namest, sizeof(namest->inumber)));
1083                 xfs_da_buf_done(bp);
1084                 blk->bp = NULL;
1085                 retval = 0;
1086         } else {
1087                 i = state->path.active - 1;
1088                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1089                 state->path.blk[i].bp = NULL;
1090         }
1091         for (i = 0; i < state->path.active - 1; i++) {
1092                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1093                 state->path.blk[i].bp = NULL;
1094         }
1095
1096         xfs_da_state_free(state);
1097         return(retval);
1098 }
1099
1100 #if defined(XFS_DIR_TRACE)
1101 /*
1102  * Add a trace buffer entry for an inode and a uio.
1103  */
1104 void
1105 xfs_dir_trace_g_du(char *where, xfs_inode_t *dp, uio_t *uio)
1106 {
1107         xfs_dir_trace_enter(XFS_DIR_KTRACE_G_DU, where,
1108                      (void *)dp, (void *)dp->i_mount,
1109                      (void *)((unsigned long)(uio->uio_offset >> 32)),
1110                      (void *)((unsigned long)(uio->uio_offset & 0xFFFFFFFF)),
1111                      (void *)(unsigned long)uio->uio_resid,
1112                      NULL, NULL, NULL, NULL, NULL, NULL, NULL);
1113 }
1114
1115 /*
1116  * Add a trace buffer entry for an inode and a uio.
1117  */
1118 void
1119 xfs_dir_trace_g_dub(char *where, xfs_inode_t *dp, uio_t *uio, xfs_dablk_t bno)
1120 {
1121         xfs_dir_trace_enter(XFS_DIR_KTRACE_G_DUB, where,
1122                      (void *)dp, (void *)dp->i_mount,
1123                      (void *)((unsigned long)(uio->uio_offset >> 32)),
1124                      (void *)((unsigned long)(uio->uio_offset & 0xFFFFFFFF)),
1125                      (void *)(unsigned long)uio->uio_resid,
1126                      (void *)(unsigned long)bno,
1127                      NULL, NULL, NULL, NULL, NULL, NULL);
1128 }
1129
1130 /*
1131  * Add a trace buffer entry for an inode and a uio.
1132  */
1133 void
1134 xfs_dir_trace_g_dun(char *where, xfs_inode_t *dp, uio_t *uio,
1135                         xfs_da_intnode_t *node)
1136 {
1137         int     last = INT_GET(node->hdr.count, ARCH_CONVERT) - 1;
1138
1139         xfs_dir_trace_enter(XFS_DIR_KTRACE_G_DUN, where,
1140                      (void *)dp, (void *)dp->i_mount,
1141                      (void *)((unsigned long)(uio->uio_offset >> 32)),
1142                      (void *)((unsigned long)(uio->uio_offset & 0xFFFFFFFF)),
1143                      (void *)(unsigned long)uio->uio_resid,
1144                      (void *)(unsigned long)
1145                         INT_GET(node->hdr.info.forw, ARCH_CONVERT),
1146                      (void *)(unsigned long)
1147                         INT_GET(node->hdr.count, ARCH_CONVERT),
1148                      (void *)(unsigned long)
1149                         INT_GET(node->btree[0].hashval, ARCH_CONVERT),
1150                      (void *)(unsigned long)
1151                         INT_GET(node->btree[last].hashval, ARCH_CONVERT),
1152                      NULL, NULL, NULL);
1153 }
1154
1155 /*
1156  * Add a trace buffer entry for an inode and a uio.
1157  */
1158 void
1159 xfs_dir_trace_g_dul(char *where, xfs_inode_t *dp, uio_t *uio,
1160                         xfs_dir_leafblock_t *leaf)
1161 {
1162         int     last = INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1;
1163
1164         xfs_dir_trace_enter(XFS_DIR_KTRACE_G_DUL, where,
1165                      (void *)dp, (void *)dp->i_mount,
1166                      (void *)((unsigned long)(uio->uio_offset >> 32)),
1167                      (void *)((unsigned long)(uio->uio_offset & 0xFFFFFFFF)),
1168                      (void *)(unsigned long)uio->uio_resid,
1169                      (void *)(unsigned long)
1170                         INT_GET(leaf->hdr.info.forw, ARCH_CONVERT),
1171                      (void *)(unsigned long)
1172                         INT_GET(leaf->hdr.count, ARCH_CONVERT),
1173                      (void *)(unsigned long)
1174                         INT_GET(leaf->entries[0].hashval, ARCH_CONVERT),
1175                      (void *)(unsigned long)
1176                         INT_GET(leaf->entries[last].hashval, ARCH_CONVERT),
1177                      NULL, NULL, NULL);
1178 }
1179
1180 /*
1181  * Add a trace buffer entry for an inode and a uio.
1182  */
1183 void
1184 xfs_dir_trace_g_due(char *where, xfs_inode_t *dp, uio_t *uio,
1185                         xfs_dir_leaf_entry_t *entry)
1186 {
1187         xfs_dir_trace_enter(XFS_DIR_KTRACE_G_DUE, where,
1188                      (void *)dp, (void *)dp->i_mount,
1189                      (void *)((unsigned long)(uio->uio_offset >> 32)),
1190                      (void *)((unsigned long)(uio->uio_offset & 0xFFFFFFFF)),
1191                      (void *)(unsigned long)uio->uio_resid,
1192                      (void *)(unsigned long)
1193                         INT_GET(entry->hashval, ARCH_CONVERT),
1194                      NULL, NULL, NULL, NULL, NULL, NULL);
1195 }
1196
1197 /*
1198  * Add a trace buffer entry for an inode and a uio.
1199  */
1200 void
1201 xfs_dir_trace_g_duc(char *where, xfs_inode_t *dp, uio_t *uio, xfs_off_t cookie)
1202 {
1203         xfs_dir_trace_enter(XFS_DIR_KTRACE_G_DUC, where,
1204                      (void *)dp, (void *)dp->i_mount,
1205                      (void *)((unsigned long)(uio->uio_offset >> 32)),
1206                      (void *)((unsigned long)(uio->uio_offset & 0xFFFFFFFF)),
1207                      (void *)(unsigned long)uio->uio_resid,
1208                      (void *)((unsigned long)(cookie >> 32)),
1209                      (void *)((unsigned long)(cookie & 0xFFFFFFFF)),
1210                      NULL, NULL, NULL, NULL, NULL);
1211 }
1212
1213 /*
1214  * Add a trace buffer entry for the arguments given to the routine,
1215  * generic form.
1216  */
1217 void
1218 xfs_dir_trace_enter(int type, char *where,
1219                         void * a0, void * a1,
1220                         void * a2, void * a3,
1221                         void * a4, void * a5,
1222                         void * a6, void * a7,
1223                         void * a8, void * a9,
1224                         void * a10, void * a11)
1225 {
1226         ASSERT(xfs_dir_trace_buf);
1227         ktrace_enter(xfs_dir_trace_buf, (void *)(unsigned long)type,
1228                                         (void *)where,
1229                                         (void *)a0, (void *)a1, (void *)a2,
1230                                         (void *)a3, (void *)a4, (void *)a5,
1231                                         (void *)a6, (void *)a7, (void *)a8,
1232                                         (void *)a9, (void *)a10, (void *)a11,
1233                                         NULL, NULL);
1234 }
1235 #endif  /* XFS_DIR_TRACE */