Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6] / fs / ocfs2 / dir.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dir.c
5  *
6  * Creates, reads, walks and deletes directory-nodes
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43
44 #define MLOG_MASK_PREFIX ML_NAMEI
45 #include <cluster/masklog.h>
46
47 #include "ocfs2.h"
48
49 #include "alloc.h"
50 #include "dir.h"
51 #include "dlmglue.h"
52 #include "extent_map.h"
53 #include "file.h"
54 #include "inode.h"
55 #include "journal.h"
56 #include "namei.h"
57 #include "suballoc.h"
58 #include "uptodate.h"
59
60 #include "buffer_head_io.h"
61
62 static unsigned char ocfs2_filetype_table[] = {
63         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
64 };
65
66 static int ocfs2_extend_dir(struct ocfs2_super *osb,
67                             struct inode *dir,
68                             struct buffer_head *parent_fe_bh,
69                             struct buffer_head **new_de_bh);
70 /*
71  * ocfs2_readdir()
72  *
73  */
74 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
75 {
76         int error = 0;
77         unsigned long offset, blk, last_ra_blk = 0;
78         int i, stored;
79         struct buffer_head * bh, * tmp;
80         struct ocfs2_dir_entry * de;
81         int err;
82         struct inode *inode = filp->f_path.dentry->d_inode;
83         struct super_block * sb = inode->i_sb;
84         unsigned int ra_sectors = 16;
85         int lock_level = 0;
86
87         mlog_entry("dirino=%llu\n",
88                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
89
90         stored = 0;
91         bh = NULL;
92
93         error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
94         if (lock_level && error >= 0) {
95                 /* We release EX lock which used to update atime
96                  * and get PR lock again to reduce contention
97                  * on commonly accessed directories. */
98                 ocfs2_meta_unlock(inode, 1);
99                 lock_level = 0;
100                 error = ocfs2_meta_lock(inode, NULL, 0);
101         }
102         if (error < 0) {
103                 if (error != -ENOENT)
104                         mlog_errno(error);
105                 /* we haven't got any yet, so propagate the error. */
106                 stored = error;
107                 goto bail_nolock;
108         }
109
110         offset = filp->f_pos & (sb->s_blocksize - 1);
111
112         while (!error && !stored && filp->f_pos < i_size_read(inode)) {
113                 blk = (filp->f_pos) >> sb->s_blocksize_bits;
114                 bh = ocfs2_bread(inode, blk, &err, 0);
115                 if (!bh) {
116                         mlog(ML_ERROR,
117                              "directory #%llu contains a hole at offset %lld\n",
118                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
119                              filp->f_pos);
120                         filp->f_pos += sb->s_blocksize - offset;
121                         continue;
122                 }
123
124                 /* The idea here is to begin with 8k read-ahead and to stay
125                  * 4k ahead of our current position.
126                  *
127                  * TODO: Use the pagecache for this. We just need to
128                  * make sure it's cluster-safe... */
129                 if (!last_ra_blk
130                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
131                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
132                              i > 0; i--) {
133                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
134                                 if (tmp)
135                                         brelse(tmp);
136                         }
137                         last_ra_blk = blk;
138                         ra_sectors = 8;
139                 }
140
141 revalidate:
142                 /* If the dir block has changed since the last call to
143                  * readdir(2), then we might be pointing to an invalid
144                  * dirent right now.  Scan from the start of the block
145                  * to make sure. */
146                 if (filp->f_version != inode->i_version) {
147                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
148                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
149                                 /* It's too expensive to do a full
150                                  * dirent test each time round this
151                                  * loop, but we do have to test at
152                                  * least that it is non-zero.  A
153                                  * failure will be detected in the
154                                  * dirent test below. */
155                                 if (le16_to_cpu(de->rec_len) <
156                                     OCFS2_DIR_REC_LEN(1))
157                                         break;
158                                 i += le16_to_cpu(de->rec_len);
159                         }
160                         offset = i;
161                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
162                                 | offset;
163                         filp->f_version = inode->i_version;
164                 }
165
166                 while (!error && filp->f_pos < i_size_read(inode)
167                        && offset < sb->s_blocksize) {
168                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
169                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
170                                 /* On error, skip the f_pos to the
171                                    next block. */
172                                 filp->f_pos = (filp->f_pos |
173                                                (sb->s_blocksize - 1)) + 1;
174                                 brelse(bh);
175                                 goto bail;
176                         }
177                         offset += le16_to_cpu(de->rec_len);
178                         if (le64_to_cpu(de->inode)) {
179                                 /* We might block in the next section
180                                  * if the data destination is
181                                  * currently swapped out.  So, use a
182                                  * version stamp to detect whether or
183                                  * not the directory has been modified
184                                  * during the copy operation.
185                                  */
186                                 unsigned long version = filp->f_version;
187                                 unsigned char d_type = DT_UNKNOWN;
188
189                                 if (de->file_type < OCFS2_FT_MAX)
190                                         d_type = ocfs2_filetype_table[de->file_type];
191                                 error = filldir(dirent, de->name,
192                                                 de->name_len,
193                                                 filp->f_pos,
194                                                 ino_from_blkno(sb, le64_to_cpu(de->inode)),
195                                                 d_type);
196                                 if (error)
197                                         break;
198                                 if (version != filp->f_version)
199                                         goto revalidate;
200                                 stored ++;
201                         }
202                         filp->f_pos += le16_to_cpu(de->rec_len);
203                 }
204                 offset = 0;
205                 brelse(bh);
206         }
207
208         stored = 0;
209 bail:
210         ocfs2_meta_unlock(inode, lock_level);
211
212 bail_nolock:
213         mlog_exit(stored);
214
215         return stored;
216 }
217
218 /*
219  * NOTE: this should always be called with parent dir i_mutex taken.
220  */
221 int ocfs2_find_files_on_disk(const char *name,
222                              int namelen,
223                              u64 *blkno,
224                              struct inode *inode,
225                              struct buffer_head **dirent_bh,
226                              struct ocfs2_dir_entry **dirent)
227 {
228         int status = -ENOENT;
229
230         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
231                    namelen, name, blkno, inode, dirent_bh, dirent);
232
233         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
234         if (!*dirent_bh || !*dirent) {
235                 status = -ENOENT;
236                 goto leave;
237         }
238
239         *blkno = le64_to_cpu((*dirent)->inode);
240
241         status = 0;
242 leave:
243         if (status < 0) {
244                 *dirent = NULL;
245                 if (*dirent_bh) {
246                         brelse(*dirent_bh);
247                         *dirent_bh = NULL;
248                 }
249         }
250
251         mlog_exit(status);
252         return status;
253 }
254
255 /* Check for a name within a directory.
256  *
257  * Return 0 if the name does not exist
258  * Return -EEXIST if the directory contains the name
259  *
260  * Callers should have i_mutex + a cluster lock on dir
261  */
262 int ocfs2_check_dir_for_entry(struct inode *dir,
263                               const char *name,
264                               int namelen)
265 {
266         int ret;
267         struct buffer_head *dirent_bh = NULL;
268         struct ocfs2_dir_entry *dirent = NULL;
269
270         mlog_entry("dir %llu, name '%.*s'\n",
271                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
272
273         ret = -EEXIST;
274         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
275         if (dirent_bh)
276                 goto bail;
277
278         ret = 0;
279 bail:
280         if (dirent_bh)
281                 brelse(dirent_bh);
282
283         mlog_exit(ret);
284         return ret;
285 }
286
287 /*
288  * routine to check that the specified directory is empty (for rmdir)
289  */
290 int ocfs2_empty_dir(struct inode *inode)
291 {
292         unsigned long offset;
293         struct buffer_head * bh;
294         struct ocfs2_dir_entry * de, * de1;
295         struct super_block * sb;
296         int err;
297
298         sb = inode->i_sb;
299         if ((i_size_read(inode) <
300              (OCFS2_DIR_REC_LEN(1) + OCFS2_DIR_REC_LEN(2))) ||
301             !(bh = ocfs2_bread(inode, 0, &err, 0))) {
302                 mlog(ML_ERROR, "bad directory (dir #%llu) - no data block\n",
303                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
304                 return 1;
305         }
306
307         de = (struct ocfs2_dir_entry *) bh->b_data;
308         de1 = (struct ocfs2_dir_entry *)
309                         ((char *)de + le16_to_cpu(de->rec_len));
310         if ((le64_to_cpu(de->inode) != OCFS2_I(inode)->ip_blkno) ||
311                         !le64_to_cpu(de1->inode) ||
312                         strcmp(".", de->name) ||
313                         strcmp("..", de1->name)) {
314                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
315                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
316                 brelse(bh);
317                 return 1;
318         }
319         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
320         de = (struct ocfs2_dir_entry *)((char *)de1 + le16_to_cpu(de1->rec_len));
321         while (offset < i_size_read(inode) ) {
322                 if (!bh || (void *)de >= (void *)(bh->b_data + sb->s_blocksize)) {
323                         brelse(bh);
324                         bh = ocfs2_bread(inode,
325                                          offset >> sb->s_blocksize_bits, &err, 0);
326                         if (!bh) {
327                                 mlog(ML_ERROR, "dir %llu has a hole at %lu\n",
328                                      (unsigned long long)OCFS2_I(inode)->ip_blkno, offset);
329                                 offset += sb->s_blocksize;
330                                 continue;
331                         }
332                         de = (struct ocfs2_dir_entry *) bh->b_data;
333                 }
334                 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
335                         brelse(bh);
336                         return 1;
337                 }
338                 if (le64_to_cpu(de->inode)) {
339                         brelse(bh);
340                         return 0;
341                 }
342                 offset += le16_to_cpu(de->rec_len);
343                 de = (struct ocfs2_dir_entry *)
344                         ((char *)de + le16_to_cpu(de->rec_len));
345         }
346         brelse(bh);
347         return 1;
348 }
349
350 /* returns a bh of the 1st new block in the allocation. */
351 int ocfs2_do_extend_dir(struct super_block *sb,
352                         handle_t *handle,
353                         struct inode *dir,
354                         struct buffer_head *parent_fe_bh,
355                         struct ocfs2_alloc_context *data_ac,
356                         struct ocfs2_alloc_context *meta_ac,
357                         struct buffer_head **new_bh)
358 {
359         int status;
360         int extend;
361         u64 p_blkno;
362
363         spin_lock(&OCFS2_I(dir)->ip_lock);
364         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
365         spin_unlock(&OCFS2_I(dir)->ip_lock);
366
367         if (extend) {
368                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, 1,
369                                                     parent_fe_bh, handle,
370                                                     data_ac, meta_ac, NULL);
371                 BUG_ON(status == -EAGAIN);
372                 if (status < 0) {
373                         mlog_errno(status);
374                         goto bail;
375                 }
376         }
377
378         status = ocfs2_extent_map_get_blocks(dir, (dir->i_blocks >>
379                                                    (sb->s_blocksize_bits - 9)),
380                                              1, &p_blkno, NULL);
381         if (status < 0) {
382                 mlog_errno(status);
383                 goto bail;
384         }
385
386         *new_bh = sb_getblk(sb, p_blkno);
387         if (!*new_bh) {
388                 status = -EIO;
389                 mlog_errno(status);
390                 goto bail;
391         }
392         status = 0;
393 bail:
394         mlog_exit(status);
395         return status;
396 }
397
398 /* assumes you already have a cluster lock on the directory. */
399 static int ocfs2_extend_dir(struct ocfs2_super *osb,
400                             struct inode *dir,
401                             struct buffer_head *parent_fe_bh,
402                             struct buffer_head **new_de_bh)
403 {
404         int status = 0;
405         int credits, num_free_extents;
406         loff_t dir_i_size;
407         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
408         struct ocfs2_alloc_context *data_ac = NULL;
409         struct ocfs2_alloc_context *meta_ac = NULL;
410         handle_t *handle = NULL;
411         struct buffer_head *new_bh = NULL;
412         struct ocfs2_dir_entry * de;
413         struct super_block *sb = osb->sb;
414
415         mlog_entry_void();
416
417         dir_i_size = i_size_read(dir);
418         mlog(0, "extending dir %llu (i_size = %lld)\n",
419              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
420
421         /* dir->i_size is always block aligned. */
422         spin_lock(&OCFS2_I(dir)->ip_lock);
423         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
424                 spin_unlock(&OCFS2_I(dir)->ip_lock);
425                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
426                 if (num_free_extents < 0) {
427                         status = num_free_extents;
428                         mlog_errno(status);
429                         goto bail;
430                 }
431
432                 if (!num_free_extents) {
433                         status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
434                         if (status < 0) {
435                                 if (status != -ENOSPC)
436                                         mlog_errno(status);
437                                 goto bail;
438                         }
439                 }
440
441                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
442                 if (status < 0) {
443                         if (status != -ENOSPC)
444                                 mlog_errno(status);
445                         goto bail;
446                 }
447
448                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
449         } else {
450                 spin_unlock(&OCFS2_I(dir)->ip_lock);
451                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
452         }
453
454         handle = ocfs2_start_trans(osb, credits);
455         if (IS_ERR(handle)) {
456                 status = PTR_ERR(handle);
457                 handle = NULL;
458                 mlog_errno(status);
459                 goto bail;
460         }
461
462         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
463                                      data_ac, meta_ac, &new_bh);
464         if (status < 0) {
465                 mlog_errno(status);
466                 goto bail;
467         }
468
469         ocfs2_set_new_buffer_uptodate(dir, new_bh);
470
471         status = ocfs2_journal_access(handle, dir, new_bh,
472                                       OCFS2_JOURNAL_ACCESS_CREATE);
473         if (status < 0) {
474                 mlog_errno(status);
475                 goto bail;
476         }
477         memset(new_bh->b_data, 0, sb->s_blocksize);
478         de = (struct ocfs2_dir_entry *) new_bh->b_data;
479         de->inode = 0;
480         de->rec_len = cpu_to_le16(sb->s_blocksize);
481         status = ocfs2_journal_dirty(handle, new_bh);
482         if (status < 0) {
483                 mlog_errno(status);
484                 goto bail;
485         }
486
487         dir_i_size += dir->i_sb->s_blocksize;
488         i_size_write(dir, dir_i_size);
489         dir->i_blocks = ocfs2_align_bytes_to_sectors(dir_i_size);
490         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
491         if (status < 0) {
492                 mlog_errno(status);
493                 goto bail;
494         }
495
496         *new_de_bh = new_bh;
497         get_bh(*new_de_bh);
498 bail:
499         if (handle)
500                 ocfs2_commit_trans(osb, handle);
501
502         if (data_ac)
503                 ocfs2_free_alloc_context(data_ac);
504         if (meta_ac)
505                 ocfs2_free_alloc_context(meta_ac);
506
507         if (new_bh)
508                 brelse(new_bh);
509
510         mlog_exit(status);
511         return status;
512 }
513
514 /*
515  * Search the dir for a good spot, extending it if necessary. The
516  * block containing an appropriate record is returned in ret_de_bh.
517  */
518 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
519                                  struct inode *dir,
520                                  struct buffer_head *parent_fe_bh,
521                                  const char *name,
522                                  int namelen,
523                                  struct buffer_head **ret_de_bh)
524 {
525         unsigned long offset;
526         struct buffer_head * bh = NULL;
527         unsigned short rec_len;
528         struct ocfs2_dinode *fe;
529         struct ocfs2_dir_entry *de;
530         struct super_block *sb;
531         int status;
532
533         mlog_entry_void();
534
535         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
536              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
537
538         BUG_ON(!S_ISDIR(dir->i_mode));
539         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
540         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
541
542         sb = dir->i_sb;
543
544         if (!namelen) {
545                 status = -EINVAL;
546                 mlog_errno(status);
547                 goto bail;
548         }
549
550         bh = ocfs2_bread(dir, 0, &status, 0);
551         if (!bh) {
552                 mlog_errno(status);
553                 goto bail;
554         }
555
556         rec_len = OCFS2_DIR_REC_LEN(namelen);
557         offset = 0;
558         de = (struct ocfs2_dir_entry *) bh->b_data;
559         while (1) {
560                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
561                         brelse(bh);
562                         bh = NULL;
563
564                         if (i_size_read(dir) <= offset) {
565                                 status = ocfs2_extend_dir(osb,
566                                                           dir,
567                                                           parent_fe_bh,
568                                                           &bh);
569                                 if (status < 0) {
570                                         mlog_errno(status);
571                                         goto bail;
572                                 }
573                                 BUG_ON(!bh);
574                                 *ret_de_bh = bh;
575                                 get_bh(*ret_de_bh);
576                                 goto bail;
577                         }
578                         bh = ocfs2_bread(dir,
579                                          offset >> sb->s_blocksize_bits,
580                                          &status,
581                                          0);
582                         if (!bh) {
583                                 mlog_errno(status);
584                                 goto bail;
585                         }
586                         /* move to next block */
587                         de = (struct ocfs2_dir_entry *) bh->b_data;
588                 }
589                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
590                         status = -ENOENT;
591                         goto bail;
592                 }
593                 if (ocfs2_match(namelen, name, de)) {
594                         status = -EEXIST;
595                         goto bail;
596                 }
597                 if (((le64_to_cpu(de->inode) == 0) &&
598                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
599                     (le16_to_cpu(de->rec_len) >=
600                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
601                         /* Ok, we found a spot. Return this bh and let
602                          * the caller actually fill it in. */
603                         *ret_de_bh = bh;
604                         get_bh(*ret_de_bh);
605                         status = 0;
606                         goto bail;
607                 }
608                 offset += le16_to_cpu(de->rec_len);
609                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
610         }
611
612         status = 0;
613 bail:
614         if (bh)
615                 brelse(bh);
616
617         mlog_exit(status);
618         return status;
619 }