]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - fs/ocfs2/dir.c
f2e2ffbf6c95f0f901ec3130e612be2d55247e69
[linux-2.6-omap-h63xx.git] / 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 "super.h"
59 #include "uptodate.h"
60
61 #include "buffer_head_io.h"
62
63 #define NAMEI_RA_CHUNKS  2
64 #define NAMEI_RA_BLOCKS  4
65 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
66 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
67
68 static unsigned char ocfs2_filetype_table[] = {
69         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
70 };
71
72 static int ocfs2_extend_dir(struct ocfs2_super *osb,
73                             struct inode *dir,
74                             struct buffer_head *parent_fe_bh,
75                             struct buffer_head **new_de_bh);
76 static int ocfs2_do_extend_dir(struct super_block *sb,
77                                handle_t *handle,
78                                struct inode *dir,
79                                struct buffer_head *parent_fe_bh,
80                                struct ocfs2_alloc_context *data_ac,
81                                struct ocfs2_alloc_context *meta_ac,
82                                struct buffer_head **new_bh);
83
84 static int ocfs2_check_dir_entry(struct inode * dir,
85                                  struct ocfs2_dir_entry * de,
86                                  struct buffer_head * bh,
87                                  unsigned long offset)
88 {
89         const char *error_msg = NULL;
90         const int rlen = le16_to_cpu(de->rec_len);
91
92         if (rlen < OCFS2_DIR_REC_LEN(1))
93                 error_msg = "rec_len is smaller than minimal";
94         else if (rlen % 4 != 0)
95                 error_msg = "rec_len % 4 != 0";
96         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
97                 error_msg = "rec_len is too small for name_len";
98         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
99                 error_msg = "directory entry across blocks";
100
101         if (error_msg != NULL)
102                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
103                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
104                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
105                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
106                      de->name_len);
107         return error_msg == NULL ? 1 : 0;
108 }
109
110 static inline int ocfs2_match(int len,
111                               const char * const name,
112                               struct ocfs2_dir_entry *de)
113 {
114         if (len != de->name_len)
115                 return 0;
116         if (!de->inode)
117                 return 0;
118         return !memcmp(name, de->name, len);
119 }
120
121 /*
122  * Returns 0 if not found, -1 on failure, and 1 on success
123  */
124 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
125                                         struct inode *dir,
126                                         const char *name, int namelen,
127                                         unsigned long offset,
128                                         struct ocfs2_dir_entry **res_dir)
129 {
130         struct ocfs2_dir_entry *de;
131         char *dlimit, *de_buf;
132         int de_len;
133         int ret = 0;
134
135         mlog_entry_void();
136
137         de_buf = bh->b_data;
138         dlimit = de_buf + dir->i_sb->s_blocksize;
139
140         while (de_buf < dlimit) {
141                 /* this code is executed quadratically often */
142                 /* do minimal checking `by hand' */
143
144                 de = (struct ocfs2_dir_entry *) de_buf;
145
146                 if (de_buf + namelen <= dlimit &&
147                     ocfs2_match(namelen, name, de)) {
148                         /* found a match - just to be sure, do a full check */
149                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
150                                 ret = -1;
151                                 goto bail;
152                         }
153                         *res_dir = de;
154                         ret = 1;
155                         goto bail;
156                 }
157
158                 /* prevent looping on a bad block */
159                 de_len = le16_to_cpu(de->rec_len);
160                 if (de_len <= 0) {
161                         ret = -1;
162                         goto bail;
163                 }
164
165                 de_buf += de_len;
166                 offset += de_len;
167         }
168
169 bail:
170         mlog_exit(ret);
171         return ret;
172 }
173
174 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
175                                      struct inode *dir,
176                                      struct ocfs2_dir_entry **res_dir)
177 {
178         struct super_block *sb;
179         struct buffer_head *bh_use[NAMEI_RA_SIZE];
180         struct buffer_head *bh, *ret = NULL;
181         unsigned long start, block, b;
182         int ra_max = 0;         /* Number of bh's in the readahead
183                                    buffer, bh_use[] */
184         int ra_ptr = 0;         /* Current index into readahead
185                                    buffer */
186         int num = 0;
187         int nblocks, i, err;
188
189         mlog_entry_void();
190
191         *res_dir = NULL;
192         sb = dir->i_sb;
193
194         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
195         start = OCFS2_I(dir)->ip_dir_start_lookup;
196         if (start >= nblocks)
197                 start = 0;
198         block = start;
199
200 restart:
201         do {
202                 /*
203                  * We deal with the read-ahead logic here.
204                  */
205                 if (ra_ptr >= ra_max) {
206                         /* Refill the readahead buffer */
207                         ra_ptr = 0;
208                         b = block;
209                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
210                                 /*
211                                  * Terminate if we reach the end of the
212                                  * directory and must wrap, or if our
213                                  * search has finished at this block.
214                                  */
215                                 if (b >= nblocks || (num && block == start)) {
216                                         bh_use[ra_max] = NULL;
217                                         break;
218                                 }
219                                 num++;
220
221                                 bh = ocfs2_bread(dir, b++, &err, 1);
222                                 bh_use[ra_max] = bh;
223                         }
224                 }
225                 if ((bh = bh_use[ra_ptr++]) == NULL)
226                         goto next;
227                 wait_on_buffer(bh);
228                 if (!buffer_uptodate(bh)) {
229                         /* read error, skip block & hope for the best */
230                         ocfs2_error(dir->i_sb, "reading directory %llu, "
231                                     "offset %lu\n",
232                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
233                                     block);
234                         brelse(bh);
235                         goto next;
236                 }
237                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
238                                           block << sb->s_blocksize_bits,
239                                           res_dir);
240                 if (i == 1) {
241                         OCFS2_I(dir)->ip_dir_start_lookup = block;
242                         ret = bh;
243                         goto cleanup_and_exit;
244                 } else {
245                         brelse(bh);
246                         if (i < 0)
247                                 goto cleanup_and_exit;
248                 }
249         next:
250                 if (++block >= nblocks)
251                         block = 0;
252         } while (block != start);
253
254         /*
255          * If the directory has grown while we were searching, then
256          * search the last part of the directory before giving up.
257          */
258         block = nblocks;
259         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
260         if (block < nblocks) {
261                 start = 0;
262                 goto restart;
263         }
264
265 cleanup_and_exit:
266         /* Clean up the read-ahead blocks */
267         for (; ra_ptr < ra_max; ra_ptr++)
268                 brelse(bh_use[ra_ptr]);
269
270         mlog_exit_ptr(ret);
271         return ret;
272 }
273
274 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
275                        struct buffer_head *de_bh, struct ocfs2_dir_entry *de,
276                        struct inode *new_entry_inode)
277 {
278         int ret;
279
280         ret = ocfs2_journal_access(handle, dir, de_bh,
281                                    OCFS2_JOURNAL_ACCESS_WRITE);
282         if (ret) {
283                 mlog_errno(ret);
284                 goto out;
285         }
286
287         de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
288         ocfs2_set_de_type(de, new_entry_inode->i_mode);
289
290         ocfs2_journal_dirty(handle, de_bh);
291
292 out:
293         return ret;
294 }
295
296 /*
297  * ocfs2_delete_entry deletes a directory entry by merging it with the
298  * previous entry
299  */
300 int ocfs2_delete_entry(handle_t *handle,
301                        struct inode *dir,
302                        struct ocfs2_dir_entry *de_del,
303                        struct buffer_head *bh)
304 {
305         struct ocfs2_dir_entry *de, *pde;
306         int i, status = -ENOENT;
307
308         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
309
310         i = 0;
311         pde = NULL;
312         de = (struct ocfs2_dir_entry *) bh->b_data;
313         while (i < bh->b_size) {
314                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
315                         status = -EIO;
316                         mlog_errno(status);
317                         goto bail;
318                 }
319                 if (de == de_del)  {
320                         status = ocfs2_journal_access(handle, dir, bh,
321                                                       OCFS2_JOURNAL_ACCESS_WRITE);
322                         if (status < 0) {
323                                 status = -EIO;
324                                 mlog_errno(status);
325                                 goto bail;
326                         }
327                         if (pde)
328                                 pde->rec_len =
329                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
330                                                     le16_to_cpu(de->rec_len));
331                         else
332                                 de->inode = 0;
333                         dir->i_version++;
334                         status = ocfs2_journal_dirty(handle, bh);
335                         goto bail;
336                 }
337                 i += le16_to_cpu(de->rec_len);
338                 pde = de;
339                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
340         }
341 bail:
342         mlog_exit(status);
343         return status;
344 }
345
346 /*
347  * Check whether 'de' has enough room to hold an entry of
348  * 'new_rec_len' bytes.
349  */
350 static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
351                                          unsigned int new_rec_len)
352 {
353         unsigned int de_really_used;
354
355         /* Check whether this is an empty record with enough space */
356         if (le64_to_cpu(de->inode) == 0 &&
357             le16_to_cpu(de->rec_len) >= new_rec_len)
358                 return 1;
359
360         /*
361          * Record might have free space at the end which we can
362          * use.
363          */
364         de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
365         if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
366             return 1;
367
368         return 0;
369 }
370
371 /* we don't always have a dentry for what we want to add, so people
372  * like orphan dir can call this instead.
373  *
374  * If you pass me insert_bh, I'll skip the search of the other dir
375  * blocks and put the record in there.
376  */
377 int __ocfs2_add_entry(handle_t *handle,
378                       struct inode *dir,
379                       const char *name, int namelen,
380                       struct inode *inode, u64 blkno,
381                       struct buffer_head *parent_fe_bh,
382                       struct buffer_head *insert_bh)
383 {
384         unsigned long offset;
385         unsigned short rec_len;
386         struct ocfs2_dir_entry *de, *de1;
387         struct super_block *sb;
388         int retval, status;
389
390         mlog_entry_void();
391
392         sb = dir->i_sb;
393
394         if (!namelen)
395                 return -EINVAL;
396
397         rec_len = OCFS2_DIR_REC_LEN(namelen);
398         offset = 0;
399         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
400         while (1) {
401                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
402                 /* These checks should've already been passed by the
403                  * prepare function, but I guess we can leave them
404                  * here anyway. */
405                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
406                         retval = -ENOENT;
407                         goto bail;
408                 }
409                 if (ocfs2_match(namelen, name, de)) {
410                         retval = -EEXIST;
411                         goto bail;
412                 }
413
414                 if (ocfs2_dirent_would_fit(de, rec_len)) {
415                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
416                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
417                         if (retval < 0) {
418                                 mlog_errno(retval);
419                                 goto bail;
420                         }
421
422                         status = ocfs2_journal_access(handle, dir, insert_bh,
423                                                       OCFS2_JOURNAL_ACCESS_WRITE);
424                         /* By now the buffer is marked for journaling */
425                         offset += le16_to_cpu(de->rec_len);
426                         if (le64_to_cpu(de->inode)) {
427                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
428                                         OCFS2_DIR_REC_LEN(de->name_len));
429                                 de1->rec_len =
430                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
431                                         OCFS2_DIR_REC_LEN(de->name_len));
432                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
433                                 de = de1;
434                         }
435                         de->file_type = OCFS2_FT_UNKNOWN;
436                         if (blkno) {
437                                 de->inode = cpu_to_le64(blkno);
438                                 ocfs2_set_de_type(de, inode->i_mode);
439                         } else
440                                 de->inode = 0;
441                         de->name_len = namelen;
442                         memcpy(de->name, name, namelen);
443
444                         dir->i_version++;
445                         status = ocfs2_journal_dirty(handle, insert_bh);
446                         retval = 0;
447                         goto bail;
448                 }
449                 offset += le16_to_cpu(de->rec_len);
450                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
451         }
452
453         /* when you think about it, the assert above should prevent us
454          * from ever getting here. */
455         retval = -ENOSPC;
456 bail:
457
458         mlog_exit(retval);
459         return retval;
460 }
461
462 static int ocfs2_dir_foreach_blk(struct inode *inode, unsigned long *f_version,
463                                  loff_t *f_pos, void *priv, filldir_t filldir)
464 {
465         int error = 0;
466         unsigned long offset, blk, last_ra_blk = 0;
467         int i, stored;
468         struct buffer_head * bh, * tmp;
469         struct ocfs2_dir_entry * de;
470         int err;
471         struct super_block * sb = inode->i_sb;
472         unsigned int ra_sectors = 16;
473
474         stored = 0;
475         bh = NULL;
476
477         offset = (*f_pos) & (sb->s_blocksize - 1);
478
479         while (!error && !stored && *f_pos < i_size_read(inode)) {
480                 blk = (*f_pos) >> sb->s_blocksize_bits;
481                 bh = ocfs2_bread(inode, blk, &err, 0);
482                 if (!bh) {
483                         mlog(ML_ERROR,
484                              "directory #%llu contains a hole at offset %lld\n",
485                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
486                              *f_pos);
487                         *f_pos += sb->s_blocksize - offset;
488                         continue;
489                 }
490
491                 /* The idea here is to begin with 8k read-ahead and to stay
492                  * 4k ahead of our current position.
493                  *
494                  * TODO: Use the pagecache for this. We just need to
495                  * make sure it's cluster-safe... */
496                 if (!last_ra_blk
497                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
498                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
499                              i > 0; i--) {
500                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
501                                 if (tmp)
502                                         brelse(tmp);
503                         }
504                         last_ra_blk = blk;
505                         ra_sectors = 8;
506                 }
507
508 revalidate:
509                 /* If the dir block has changed since the last call to
510                  * readdir(2), then we might be pointing to an invalid
511                  * dirent right now.  Scan from the start of the block
512                  * to make sure. */
513                 if (*f_version != inode->i_version) {
514                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
515                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
516                                 /* It's too expensive to do a full
517                                  * dirent test each time round this
518                                  * loop, but we do have to test at
519                                  * least that it is non-zero.  A
520                                  * failure will be detected in the
521                                  * dirent test below. */
522                                 if (le16_to_cpu(de->rec_len) <
523                                     OCFS2_DIR_REC_LEN(1))
524                                         break;
525                                 i += le16_to_cpu(de->rec_len);
526                         }
527                         offset = i;
528                         *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
529                                 | offset;
530                         *f_version = inode->i_version;
531                 }
532
533                 while (!error && *f_pos < i_size_read(inode)
534                        && offset < sb->s_blocksize) {
535                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
536                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
537                                 /* On error, skip the f_pos to the
538                                    next block. */
539                                 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
540                                 brelse(bh);
541                                 goto out;
542                         }
543                         offset += le16_to_cpu(de->rec_len);
544                         if (le64_to_cpu(de->inode)) {
545                                 /* We might block in the next section
546                                  * if the data destination is
547                                  * currently swapped out.  So, use a
548                                  * version stamp to detect whether or
549                                  * not the directory has been modified
550                                  * during the copy operation.
551                                  */
552                                 unsigned long version = *f_version;
553                                 unsigned char d_type = DT_UNKNOWN;
554
555                                 if (de->file_type < OCFS2_FT_MAX)
556                                         d_type = ocfs2_filetype_table[de->file_type];
557                                 error = filldir(priv, de->name,
558                                                 de->name_len,
559                                                 *f_pos,
560                                                 le64_to_cpu(de->inode),
561                                                 d_type);
562                                 if (error)
563                                         break;
564                                 if (version != *f_version)
565                                         goto revalidate;
566                                 stored ++;
567                         }
568                         *f_pos += le16_to_cpu(de->rec_len);
569                 }
570                 offset = 0;
571                 brelse(bh);
572         }
573
574         stored = 0;
575 out:
576         return stored;
577 }
578
579 /*
580  * This is intended to be called from inside other kernel functions,
581  * so we fake some arguments.
582  */
583 int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
584                       filldir_t filldir)
585 {
586         int ret = 0;
587         unsigned long version = inode->i_version;
588
589         while (*f_pos < i_size_read(inode)) {
590                 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
591                                             filldir);
592                 if (ret)
593                         break;
594         }
595
596         return 0;
597 }
598
599 /*
600  * ocfs2_readdir()
601  *
602  */
603 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
604 {
605         int error = 0;
606         struct inode *inode = filp->f_path.dentry->d_inode;
607         int lock_level = 0;
608
609         mlog_entry("dirino=%llu\n",
610                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
611
612         error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
613         if (lock_level && error >= 0) {
614                 /* We release EX lock which used to update atime
615                  * and get PR lock again to reduce contention
616                  * on commonly accessed directories. */
617                 ocfs2_meta_unlock(inode, 1);
618                 lock_level = 0;
619                 error = ocfs2_meta_lock(inode, NULL, 0);
620         }
621         if (error < 0) {
622                 if (error != -ENOENT)
623                         mlog_errno(error);
624                 /* we haven't got any yet, so propagate the error. */
625                 goto bail_nolock;
626         }
627
628         error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
629                                       dirent, filldir);
630
631         ocfs2_meta_unlock(inode, lock_level);
632
633 bail_nolock:
634         mlog_exit(error);
635
636         return error;
637 }
638
639 /*
640  * NOTE: this should always be called with parent dir i_mutex taken.
641  */
642 int ocfs2_find_files_on_disk(const char *name,
643                              int namelen,
644                              u64 *blkno,
645                              struct inode *inode,
646                              struct buffer_head **dirent_bh,
647                              struct ocfs2_dir_entry **dirent)
648 {
649         int status = -ENOENT;
650
651         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
652                    namelen, name, blkno, inode, dirent_bh, dirent);
653
654         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
655         if (!*dirent_bh || !*dirent) {
656                 status = -ENOENT;
657                 goto leave;
658         }
659
660         *blkno = le64_to_cpu((*dirent)->inode);
661
662         status = 0;
663 leave:
664         if (status < 0) {
665                 *dirent = NULL;
666                 if (*dirent_bh) {
667                         brelse(*dirent_bh);
668                         *dirent_bh = NULL;
669                 }
670         }
671
672         mlog_exit(status);
673         return status;
674 }
675
676 /*
677  * Convenience function for callers which just want the block number
678  * mapped to a name and don't require the full dirent info, etc.
679  */
680 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
681                                int namelen, u64 *blkno)
682 {
683         int ret;
684         struct buffer_head *bh = NULL;
685         struct ocfs2_dir_entry *dirent = NULL;
686
687         ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &bh, &dirent);
688         brelse(bh);
689
690         return ret;
691 }
692
693 /* Check for a name within a directory.
694  *
695  * Return 0 if the name does not exist
696  * Return -EEXIST if the directory contains the name
697  *
698  * Callers should have i_mutex + a cluster lock on dir
699  */
700 int ocfs2_check_dir_for_entry(struct inode *dir,
701                               const char *name,
702                               int namelen)
703 {
704         int ret;
705         struct buffer_head *dirent_bh = NULL;
706         struct ocfs2_dir_entry *dirent = NULL;
707
708         mlog_entry("dir %llu, name '%.*s'\n",
709                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
710
711         ret = -EEXIST;
712         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
713         if (dirent_bh)
714                 goto bail;
715
716         ret = 0;
717 bail:
718         if (dirent_bh)
719                 brelse(dirent_bh);
720
721         mlog_exit(ret);
722         return ret;
723 }
724
725 struct ocfs2_empty_dir_priv {
726         unsigned seen_dot;
727         unsigned seen_dot_dot;
728         unsigned seen_other;
729 };
730 static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
731                                    loff_t pos, u64 ino, unsigned type)
732 {
733         struct ocfs2_empty_dir_priv *p = priv;
734
735         /*
736          * Check the positions of "." and ".." records to be sure
737          * they're in the correct place.
738          */
739         if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
740                 p->seen_dot = 1;
741                 return 0;
742         }
743
744         if (name_len == 2 && !strncmp("..", name, 2) &&
745             pos == OCFS2_DIR_REC_LEN(1)) {
746                 p->seen_dot_dot = 1;
747                 return 0;
748         }
749
750         p->seen_other = 1;
751         return 1;
752 }
753 /*
754  * routine to check that the specified directory is empty (for rmdir)
755  *
756  * Returns 1 if dir is empty, zero otherwise.
757  */
758 int ocfs2_empty_dir(struct inode *inode)
759 {
760         int ret;
761         loff_t start = 0;
762         struct ocfs2_empty_dir_priv priv;
763
764         memset(&priv, 0, sizeof(priv));
765
766         ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
767         if (ret)
768                 mlog_errno(ret);
769
770         if (!priv.seen_dot || !priv.seen_dot_dot) {
771                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
772                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
773                 /*
774                  * XXX: Is it really safe to allow an unlink to continue?
775                  */
776                 return 1;
777         }
778
779         return !priv.seen_other;
780 }
781
782 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
783                        handle_t *handle,
784                        struct inode *parent,
785                        struct inode *inode,
786                        struct buffer_head *fe_bh,
787                        struct ocfs2_alloc_context *data_ac)
788 {
789         int status;
790         struct buffer_head *new_bh = NULL;
791         struct ocfs2_dir_entry *de = NULL;
792
793         mlog_entry_void();
794
795         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
796                                      data_ac, NULL, &new_bh);
797         if (status < 0) {
798                 mlog_errno(status);
799                 goto bail;
800         }
801
802         ocfs2_set_new_buffer_uptodate(inode, new_bh);
803
804         status = ocfs2_journal_access(handle, inode, new_bh,
805                                       OCFS2_JOURNAL_ACCESS_CREATE);
806         if (status < 0) {
807                 mlog_errno(status);
808                 goto bail;
809         }
810         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
811
812         de = (struct ocfs2_dir_entry *) new_bh->b_data;
813         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
814         de->name_len = 1;
815         de->rec_len =
816                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
817         strcpy(de->name, ".");
818         ocfs2_set_de_type(de, S_IFDIR);
819         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
820         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
821         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
822                                   OCFS2_DIR_REC_LEN(1));
823         de->name_len = 2;
824         strcpy(de->name, "..");
825         ocfs2_set_de_type(de, S_IFDIR);
826
827         status = ocfs2_journal_dirty(handle, new_bh);
828         if (status < 0) {
829                 mlog_errno(status);
830                 goto bail;
831         }
832
833         i_size_write(inode, inode->i_sb->s_blocksize);
834         inode->i_nlink = 2;
835         inode->i_blocks = ocfs2_inode_sector_count(inode);
836         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
837         if (status < 0) {
838                 mlog_errno(status);
839                 goto bail;
840         }
841
842         status = 0;
843 bail:
844         if (new_bh)
845                 brelse(new_bh);
846
847         mlog_exit(status);
848         return status;
849 }
850
851 /* returns a bh of the 1st new block in the allocation. */
852 static int ocfs2_do_extend_dir(struct super_block *sb,
853                                handle_t *handle,
854                                struct inode *dir,
855                                struct buffer_head *parent_fe_bh,
856                                struct ocfs2_alloc_context *data_ac,
857                                struct ocfs2_alloc_context *meta_ac,
858                                struct buffer_head **new_bh)
859 {
860         int status;
861         int extend;
862         u64 p_blkno, v_blkno;
863
864         spin_lock(&OCFS2_I(dir)->ip_lock);
865         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
866         spin_unlock(&OCFS2_I(dir)->ip_lock);
867
868         if (extend) {
869                 u32 offset = OCFS2_I(dir)->ip_clusters;
870
871                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
872                                                     1, 0, parent_fe_bh, handle,
873                                                     data_ac, meta_ac, NULL);
874                 BUG_ON(status == -EAGAIN);
875                 if (status < 0) {
876                         mlog_errno(status);
877                         goto bail;
878                 }
879         }
880
881         v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
882         status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
883         if (status < 0) {
884                 mlog_errno(status);
885                 goto bail;
886         }
887
888         *new_bh = sb_getblk(sb, p_blkno);
889         if (!*new_bh) {
890                 status = -EIO;
891                 mlog_errno(status);
892                 goto bail;
893         }
894         status = 0;
895 bail:
896         mlog_exit(status);
897         return status;
898 }
899
900 /* assumes you already have a cluster lock on the directory. */
901 static int ocfs2_extend_dir(struct ocfs2_super *osb,
902                             struct inode *dir,
903                             struct buffer_head *parent_fe_bh,
904                             struct buffer_head **new_de_bh)
905 {
906         int status = 0;
907         int credits, num_free_extents, drop_alloc_sem = 0;
908         loff_t dir_i_size;
909         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
910         struct ocfs2_alloc_context *data_ac = NULL;
911         struct ocfs2_alloc_context *meta_ac = NULL;
912         handle_t *handle = NULL;
913         struct buffer_head *new_bh = NULL;
914         struct ocfs2_dir_entry * de;
915         struct super_block *sb = osb->sb;
916
917         mlog_entry_void();
918
919         dir_i_size = i_size_read(dir);
920         mlog(0, "extending dir %llu (i_size = %lld)\n",
921              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
922
923         /* dir->i_size is always block aligned. */
924         spin_lock(&OCFS2_I(dir)->ip_lock);
925         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
926                 spin_unlock(&OCFS2_I(dir)->ip_lock);
927                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
928                 if (num_free_extents < 0) {
929                         status = num_free_extents;
930                         mlog_errno(status);
931                         goto bail;
932                 }
933
934                 if (!num_free_extents) {
935                         status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
936                         if (status < 0) {
937                                 if (status != -ENOSPC)
938                                         mlog_errno(status);
939                                 goto bail;
940                         }
941                 }
942
943                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
944                 if (status < 0) {
945                         if (status != -ENOSPC)
946                                 mlog_errno(status);
947                         goto bail;
948                 }
949
950                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
951         } else {
952                 spin_unlock(&OCFS2_I(dir)->ip_lock);
953                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
954         }
955
956         down_write(&OCFS2_I(dir)->ip_alloc_sem);
957         drop_alloc_sem = 1;
958
959         handle = ocfs2_start_trans(osb, credits);
960         if (IS_ERR(handle)) {
961                 status = PTR_ERR(handle);
962                 handle = NULL;
963                 mlog_errno(status);
964                 goto bail;
965         }
966
967         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
968                                      data_ac, meta_ac, &new_bh);
969         if (status < 0) {
970                 mlog_errno(status);
971                 goto bail;
972         }
973
974         ocfs2_set_new_buffer_uptodate(dir, new_bh);
975
976         status = ocfs2_journal_access(handle, dir, new_bh,
977                                       OCFS2_JOURNAL_ACCESS_CREATE);
978         if (status < 0) {
979                 mlog_errno(status);
980                 goto bail;
981         }
982         memset(new_bh->b_data, 0, sb->s_blocksize);
983         de = (struct ocfs2_dir_entry *) new_bh->b_data;
984         de->inode = 0;
985         de->rec_len = cpu_to_le16(sb->s_blocksize);
986         status = ocfs2_journal_dirty(handle, new_bh);
987         if (status < 0) {
988                 mlog_errno(status);
989                 goto bail;
990         }
991
992         dir_i_size += dir->i_sb->s_blocksize;
993         i_size_write(dir, dir_i_size);
994         dir->i_blocks = ocfs2_inode_sector_count(dir);
995         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
996         if (status < 0) {
997                 mlog_errno(status);
998                 goto bail;
999         }
1000
1001         *new_de_bh = new_bh;
1002         get_bh(*new_de_bh);
1003 bail:
1004         if (drop_alloc_sem)
1005                 up_write(&OCFS2_I(dir)->ip_alloc_sem);
1006         if (handle)
1007                 ocfs2_commit_trans(osb, handle);
1008
1009         if (data_ac)
1010                 ocfs2_free_alloc_context(data_ac);
1011         if (meta_ac)
1012                 ocfs2_free_alloc_context(meta_ac);
1013
1014         if (new_bh)
1015                 brelse(new_bh);
1016
1017         mlog_exit(status);
1018         return status;
1019 }
1020
1021 /*
1022  * Search the dir for a good spot, extending it if necessary. The
1023  * block containing an appropriate record is returned in ret_de_bh.
1024  */
1025 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
1026                                  struct inode *dir,
1027                                  struct buffer_head *parent_fe_bh,
1028                                  const char *name,
1029                                  int namelen,
1030                                  struct buffer_head **ret_de_bh)
1031 {
1032         unsigned long offset;
1033         struct buffer_head * bh = NULL;
1034         unsigned short rec_len;
1035         struct ocfs2_dinode *fe;
1036         struct ocfs2_dir_entry *de;
1037         struct super_block *sb;
1038         int status;
1039
1040         mlog_entry_void();
1041
1042         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
1043              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
1044
1045         BUG_ON(!S_ISDIR(dir->i_mode));
1046         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1047         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
1048
1049         sb = dir->i_sb;
1050
1051         if (!namelen) {
1052                 status = -EINVAL;
1053                 mlog_errno(status);
1054                 goto bail;
1055         }
1056
1057         bh = ocfs2_bread(dir, 0, &status, 0);
1058         if (!bh) {
1059                 mlog_errno(status);
1060                 goto bail;
1061         }
1062
1063         rec_len = OCFS2_DIR_REC_LEN(namelen);
1064         offset = 0;
1065         de = (struct ocfs2_dir_entry *) bh->b_data;
1066         while (1) {
1067                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
1068                         brelse(bh);
1069                         bh = NULL;
1070
1071                         if (i_size_read(dir) <= offset) {
1072                                 status = ocfs2_extend_dir(osb,
1073                                                           dir,
1074                                                           parent_fe_bh,
1075                                                           &bh);
1076                                 if (status < 0) {
1077                                         mlog_errno(status);
1078                                         goto bail;
1079                                 }
1080                                 BUG_ON(!bh);
1081                                 *ret_de_bh = bh;
1082                                 get_bh(*ret_de_bh);
1083                                 goto bail;
1084                         }
1085                         bh = ocfs2_bread(dir,
1086                                          offset >> sb->s_blocksize_bits,
1087                                          &status,
1088                                          0);
1089                         if (!bh) {
1090                                 mlog_errno(status);
1091                                 goto bail;
1092                         }
1093                         /* move to next block */
1094                         de = (struct ocfs2_dir_entry *) bh->b_data;
1095                 }
1096                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1097                         status = -ENOENT;
1098                         goto bail;
1099                 }
1100                 if (ocfs2_match(namelen, name, de)) {
1101                         status = -EEXIST;
1102                         goto bail;
1103                 }
1104                 if (ocfs2_dirent_would_fit(de, rec_len)) {
1105                         /* Ok, we found a spot. Return this bh and let
1106                          * the caller actually fill it in. */
1107                         *ret_de_bh = bh;
1108                         get_bh(*ret_de_bh);
1109                         status = 0;
1110                         goto bail;
1111                 }
1112                 offset += le16_to_cpu(de->rec_len);
1113                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
1114         }
1115
1116         status = 0;
1117 bail:
1118         if (bh)
1119                 brelse(bh);
1120
1121         mlog_exit(status);
1122         return status;
1123 }