]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - fs/ocfs2/dir.c
ocfs2: Write support for directories with inline data
[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                             unsigned int blocks_wanted,
76                             struct buffer_head **new_de_bh);
77 static int ocfs2_do_extend_dir(struct super_block *sb,
78                                handle_t *handle,
79                                struct inode *dir,
80                                struct buffer_head *parent_fe_bh,
81                                struct ocfs2_alloc_context *data_ac,
82                                struct ocfs2_alloc_context *meta_ac,
83                                struct buffer_head **new_bh);
84
85 /*
86  * bh passed here can be an inode block or a dir data block, depending
87  * on the inode inline data flag.
88  */
89 static int ocfs2_check_dir_entry(struct inode * dir,
90                                  struct ocfs2_dir_entry * de,
91                                  struct buffer_head * bh,
92                                  unsigned long offset)
93 {
94         const char *error_msg = NULL;
95         const int rlen = le16_to_cpu(de->rec_len);
96
97         if (rlen < OCFS2_DIR_REC_LEN(1))
98                 error_msg = "rec_len is smaller than minimal";
99         else if (rlen % 4 != 0)
100                 error_msg = "rec_len % 4 != 0";
101         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
102                 error_msg = "rec_len is too small for name_len";
103         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
104                 error_msg = "directory entry across blocks";
105
106         if (error_msg != NULL)
107                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
108                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
109                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
110                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
111                      de->name_len);
112         return error_msg == NULL ? 1 : 0;
113 }
114
115 static inline int ocfs2_match(int len,
116                               const char * const name,
117                               struct ocfs2_dir_entry *de)
118 {
119         if (len != de->name_len)
120                 return 0;
121         if (!de->inode)
122                 return 0;
123         return !memcmp(name, de->name, len);
124 }
125
126 /*
127  * Returns 0 if not found, -1 on failure, and 1 on success
128  */
129 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
130                                         struct inode *dir,
131                                         const char *name, int namelen,
132                                         unsigned long offset,
133                                         char *first_de,
134                                         unsigned int bytes,
135                                         struct ocfs2_dir_entry **res_dir)
136 {
137         struct ocfs2_dir_entry *de;
138         char *dlimit, *de_buf;
139         int de_len;
140         int ret = 0;
141
142         mlog_entry_void();
143
144         de_buf = first_de;
145         dlimit = de_buf + bytes;
146
147         while (de_buf < dlimit) {
148                 /* this code is executed quadratically often */
149                 /* do minimal checking `by hand' */
150
151                 de = (struct ocfs2_dir_entry *) de_buf;
152
153                 if (de_buf + namelen <= dlimit &&
154                     ocfs2_match(namelen, name, de)) {
155                         /* found a match - just to be sure, do a full check */
156                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
157                                 ret = -1;
158                                 goto bail;
159                         }
160                         *res_dir = de;
161                         ret = 1;
162                         goto bail;
163                 }
164
165                 /* prevent looping on a bad block */
166                 de_len = le16_to_cpu(de->rec_len);
167                 if (de_len <= 0) {
168                         ret = -1;
169                         goto bail;
170                 }
171
172                 de_buf += de_len;
173                 offset += de_len;
174         }
175
176 bail:
177         mlog_exit(ret);
178         return ret;
179 }
180
181 static struct buffer_head *ocfs2_find_entry_id(const char *name,
182                                                int namelen,
183                                                struct inode *dir,
184                                                struct ocfs2_dir_entry **res_dir)
185 {
186         int ret, found;
187         struct buffer_head *di_bh = NULL;
188         struct ocfs2_dinode *di;
189         struct ocfs2_inline_data *data;
190
191         ret = ocfs2_read_block(OCFS2_SB(dir->i_sb), OCFS2_I(dir)->ip_blkno,
192                                &di_bh, OCFS2_BH_CACHED, dir);
193         if (ret) {
194                 mlog_errno(ret);
195                 goto out;
196         }
197
198         di = (struct ocfs2_dinode *)di_bh->b_data;
199         data = &di->id2.i_data;
200
201         found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
202                                       data->id_data, i_size_read(dir), res_dir);
203         if (found == 1)
204                 return di_bh;
205
206         brelse(di_bh);
207 out:
208         return NULL;
209 }
210
211 struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
212                                         struct inode *dir,
213                                         struct ocfs2_dir_entry **res_dir)
214 {
215         struct super_block *sb;
216         struct buffer_head *bh_use[NAMEI_RA_SIZE];
217         struct buffer_head *bh, *ret = NULL;
218         unsigned long start, block, b;
219         int ra_max = 0;         /* Number of bh's in the readahead
220                                    buffer, bh_use[] */
221         int ra_ptr = 0;         /* Current index into readahead
222                                    buffer */
223         int num = 0;
224         int nblocks, i, err;
225
226         mlog_entry_void();
227
228         sb = dir->i_sb;
229
230         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
231         start = OCFS2_I(dir)->ip_dir_start_lookup;
232         if (start >= nblocks)
233                 start = 0;
234         block = start;
235
236 restart:
237         do {
238                 /*
239                  * We deal with the read-ahead logic here.
240                  */
241                 if (ra_ptr >= ra_max) {
242                         /* Refill the readahead buffer */
243                         ra_ptr = 0;
244                         b = block;
245                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
246                                 /*
247                                  * Terminate if we reach the end of the
248                                  * directory and must wrap, or if our
249                                  * search has finished at this block.
250                                  */
251                                 if (b >= nblocks || (num && block == start)) {
252                                         bh_use[ra_max] = NULL;
253                                         break;
254                                 }
255                                 num++;
256
257                                 bh = ocfs2_bread(dir, b++, &err, 1);
258                                 bh_use[ra_max] = bh;
259                         }
260                 }
261                 if ((bh = bh_use[ra_ptr++]) == NULL)
262                         goto next;
263                 wait_on_buffer(bh);
264                 if (!buffer_uptodate(bh)) {
265                         /* read error, skip block & hope for the best */
266                         ocfs2_error(dir->i_sb, "reading directory %llu, "
267                                     "offset %lu\n",
268                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
269                                     block);
270                         brelse(bh);
271                         goto next;
272                 }
273                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
274                                           block << sb->s_blocksize_bits,
275                                           bh->b_data, sb->s_blocksize,
276                                           res_dir);
277                 if (i == 1) {
278                         OCFS2_I(dir)->ip_dir_start_lookup = block;
279                         ret = bh;
280                         goto cleanup_and_exit;
281                 } else {
282                         brelse(bh);
283                         if (i < 0)
284                                 goto cleanup_and_exit;
285                 }
286         next:
287                 if (++block >= nblocks)
288                         block = 0;
289         } while (block != start);
290
291         /*
292          * If the directory has grown while we were searching, then
293          * search the last part of the directory before giving up.
294          */
295         block = nblocks;
296         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
297         if (block < nblocks) {
298                 start = 0;
299                 goto restart;
300         }
301
302 cleanup_and_exit:
303         /* Clean up the read-ahead blocks */
304         for (; ra_ptr < ra_max; ra_ptr++)
305                 brelse(bh_use[ra_ptr]);
306
307         mlog_exit_ptr(ret);
308         return ret;
309 }
310
311 /*
312  * Try to find an entry of the provided name within 'dir'.
313  *
314  * If nothing was found, NULL is returned. Otherwise, a buffer_head
315  * and pointer to the dir entry are passed back.
316  *
317  * Caller can NOT assume anything about the contents of the
318  * buffer_head - it is passed back only so that it can be passed into
319  * any one of the manipulation functions (add entry, delete entry,
320  * etc). As an example, bh in the extent directory case is a data
321  * block, in the inline-data case it actually points to an inode.
322  */
323 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
324                                      struct inode *dir,
325                                      struct ocfs2_dir_entry **res_dir)
326 {
327         *res_dir = NULL;
328
329         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
330                 return ocfs2_find_entry_id(name, namelen, dir, res_dir);
331
332         return ocfs2_find_entry_el(name, namelen, dir, res_dir);
333 }
334
335 /*
336  * Update inode number and type of a previously found directory entry.
337  */
338 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
339                        struct buffer_head *de_bh, struct ocfs2_dir_entry *de,
340                        struct inode *new_entry_inode)
341 {
342         int ret;
343
344         /*
345          * The same code works fine for both inline-data and extent
346          * based directories, so no need to split this up.
347          */
348
349         ret = ocfs2_journal_access(handle, dir, de_bh,
350                                    OCFS2_JOURNAL_ACCESS_WRITE);
351         if (ret) {
352                 mlog_errno(ret);
353                 goto out;
354         }
355
356         de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
357         ocfs2_set_de_type(de, new_entry_inode->i_mode);
358
359         ocfs2_journal_dirty(handle, de_bh);
360
361 out:
362         return ret;
363 }
364
365 static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
366                                 struct ocfs2_dir_entry *de_del,
367                                 struct buffer_head *bh, char *first_de,
368                                 unsigned int bytes)
369 {
370         struct ocfs2_dir_entry *de, *pde;
371         int i, status = -ENOENT;
372
373         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
374
375         i = 0;
376         pde = NULL;
377         de = (struct ocfs2_dir_entry *) first_de;
378         while (i < bytes) {
379                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
380                         status = -EIO;
381                         mlog_errno(status);
382                         goto bail;
383                 }
384                 if (de == de_del)  {
385                         status = ocfs2_journal_access(handle, dir, bh,
386                                                       OCFS2_JOURNAL_ACCESS_WRITE);
387                         if (status < 0) {
388                                 status = -EIO;
389                                 mlog_errno(status);
390                                 goto bail;
391                         }
392                         if (pde)
393                                 pde->rec_len =
394                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
395                                                     le16_to_cpu(de->rec_len));
396                         else
397                                 de->inode = 0;
398                         dir->i_version++;
399                         status = ocfs2_journal_dirty(handle, bh);
400                         goto bail;
401                 }
402                 i += le16_to_cpu(de->rec_len);
403                 pde = de;
404                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
405         }
406 bail:
407         mlog_exit(status);
408         return status;
409 }
410
411 static inline int ocfs2_delete_entry_id(handle_t *handle,
412                                         struct inode *dir,
413                                         struct ocfs2_dir_entry *de_del,
414                                         struct buffer_head *bh)
415 {
416         int ret;
417         struct buffer_head *di_bh = NULL;
418         struct ocfs2_dinode *di;
419         struct ocfs2_inline_data *data;
420
421         ret = ocfs2_read_block(OCFS2_SB(dir->i_sb), OCFS2_I(dir)->ip_blkno,
422                                &di_bh, OCFS2_BH_CACHED, dir);
423         if (ret) {
424                 mlog_errno(ret);
425                 goto out;
426         }
427
428         di = (struct ocfs2_dinode *)di_bh->b_data;
429         data = &di->id2.i_data;
430
431         ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
432                                    i_size_read(dir));
433
434         brelse(di_bh);
435 out:
436         return ret;
437 }
438
439 static inline int ocfs2_delete_entry_el(handle_t *handle,
440                                         struct inode *dir,
441                                         struct ocfs2_dir_entry *de_del,
442                                         struct buffer_head *bh)
443 {
444         return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
445                                     bh->b_size);
446 }
447
448 /*
449  * ocfs2_delete_entry deletes a directory entry by merging it with the
450  * previous entry
451  */
452 int ocfs2_delete_entry(handle_t *handle,
453                        struct inode *dir,
454                        struct ocfs2_dir_entry *de_del,
455                        struct buffer_head *bh)
456 {
457         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
458                 return ocfs2_delete_entry_id(handle, dir, de_del, bh);
459
460         return ocfs2_delete_entry_el(handle, dir, de_del, bh);
461 }
462
463 /*
464  * Check whether 'de' has enough room to hold an entry of
465  * 'new_rec_len' bytes.
466  */
467 static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
468                                          unsigned int new_rec_len)
469 {
470         unsigned int de_really_used;
471
472         /* Check whether this is an empty record with enough space */
473         if (le64_to_cpu(de->inode) == 0 &&
474             le16_to_cpu(de->rec_len) >= new_rec_len)
475                 return 1;
476
477         /*
478          * Record might have free space at the end which we can
479          * use.
480          */
481         de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
482         if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
483             return 1;
484
485         return 0;
486 }
487
488 /* we don't always have a dentry for what we want to add, so people
489  * like orphan dir can call this instead.
490  *
491  * If you pass me insert_bh, I'll skip the search of the other dir
492  * blocks and put the record in there.
493  */
494 int __ocfs2_add_entry(handle_t *handle,
495                       struct inode *dir,
496                       const char *name, int namelen,
497                       struct inode *inode, u64 blkno,
498                       struct buffer_head *parent_fe_bh,
499                       struct buffer_head *insert_bh)
500 {
501         unsigned long offset;
502         unsigned short rec_len;
503         struct ocfs2_dir_entry *de, *de1;
504         struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
505         struct super_block *sb = dir->i_sb;
506         int retval, status;
507         unsigned int size = sb->s_blocksize;
508         char *data_start = insert_bh->b_data;
509
510         mlog_entry_void();
511
512         if (!namelen)
513                 return -EINVAL;
514
515         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
516                 data_start = di->id2.i_data.id_data;
517                 size = i_size_read(dir);
518
519                 BUG_ON(insert_bh != parent_fe_bh);
520         }
521
522         rec_len = OCFS2_DIR_REC_LEN(namelen);
523         offset = 0;
524         de = (struct ocfs2_dir_entry *) data_start;
525         while (1) {
526                 BUG_ON((char *)de >= (size + data_start));
527
528                 /* These checks should've already been passed by the
529                  * prepare function, but I guess we can leave them
530                  * here anyway. */
531                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
532                         retval = -ENOENT;
533                         goto bail;
534                 }
535                 if (ocfs2_match(namelen, name, de)) {
536                         retval = -EEXIST;
537                         goto bail;
538                 }
539
540                 if (ocfs2_dirent_would_fit(de, rec_len)) {
541                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
542                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
543                         if (retval < 0) {
544                                 mlog_errno(retval);
545                                 goto bail;
546                         }
547
548                         status = ocfs2_journal_access(handle, dir, insert_bh,
549                                                       OCFS2_JOURNAL_ACCESS_WRITE);
550                         /* By now the buffer is marked for journaling */
551                         offset += le16_to_cpu(de->rec_len);
552                         if (le64_to_cpu(de->inode)) {
553                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
554                                         OCFS2_DIR_REC_LEN(de->name_len));
555                                 de1->rec_len =
556                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
557                                         OCFS2_DIR_REC_LEN(de->name_len));
558                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
559                                 de = de1;
560                         }
561                         de->file_type = OCFS2_FT_UNKNOWN;
562                         if (blkno) {
563                                 de->inode = cpu_to_le64(blkno);
564                                 ocfs2_set_de_type(de, inode->i_mode);
565                         } else
566                                 de->inode = 0;
567                         de->name_len = namelen;
568                         memcpy(de->name, name, namelen);
569
570                         dir->i_version++;
571                         status = ocfs2_journal_dirty(handle, insert_bh);
572                         retval = 0;
573                         goto bail;
574                 }
575                 offset += le16_to_cpu(de->rec_len);
576                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
577         }
578
579         /* when you think about it, the assert above should prevent us
580          * from ever getting here. */
581         retval = -ENOSPC;
582 bail:
583
584         mlog_exit(retval);
585         return retval;
586 }
587
588 static int ocfs2_dir_foreach_blk_id(struct inode *inode,
589                                     unsigned long *f_version,
590                                     loff_t *f_pos, void *priv,
591                                     filldir_t filldir)
592 {
593         int ret, i, filldir_ret;
594         unsigned long offset = *f_pos;
595         struct buffer_head *di_bh = NULL;
596         struct ocfs2_dinode *di;
597         struct ocfs2_inline_data *data;
598         struct ocfs2_dir_entry *de;
599
600         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), OCFS2_I(inode)->ip_blkno,
601                                &di_bh, OCFS2_BH_CACHED, inode);
602         if (ret) {
603                 mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
604                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
605                 goto out;
606         }
607
608         di = (struct ocfs2_dinode *)di_bh->b_data;
609         data = &di->id2.i_data;
610
611         while (*f_pos < i_size_read(inode)) {
612 revalidate:
613                 /* If the dir block has changed since the last call to
614                  * readdir(2), then we might be pointing to an invalid
615                  * dirent right now.  Scan from the start of the block
616                  * to make sure. */
617                 if (*f_version != inode->i_version) {
618                         for (i = 0; i < i_size_read(inode) && i < offset; ) {
619                                 de = (struct ocfs2_dir_entry *)
620                                         (data->id_data + i);
621                                 /* It's too expensive to do a full
622                                  * dirent test each time round this
623                                  * loop, but we do have to test at
624                                  * least that it is non-zero.  A
625                                  * failure will be detected in the
626                                  * dirent test below. */
627                                 if (le16_to_cpu(de->rec_len) <
628                                     OCFS2_DIR_REC_LEN(1))
629                                         break;
630                                 i += le16_to_cpu(de->rec_len);
631                         }
632                         *f_pos = offset = i;
633                         *f_version = inode->i_version;
634                 }
635
636                 de = (struct ocfs2_dir_entry *) (data->id_data + *f_pos);
637                 if (!ocfs2_check_dir_entry(inode, de, di_bh, *f_pos)) {
638                         /* On error, skip the f_pos to the end. */
639                         *f_pos = i_size_read(inode);
640                         goto out;
641                 }
642                 offset += le16_to_cpu(de->rec_len);
643                 if (le64_to_cpu(de->inode)) {
644                         /* We might block in the next section
645                          * if the data destination is
646                          * currently swapped out.  So, use a
647                          * version stamp to detect whether or
648                          * not the directory has been modified
649                          * during the copy operation.
650                          */
651                         unsigned long version = *f_version;
652                         unsigned char d_type = DT_UNKNOWN;
653
654                         if (de->file_type < OCFS2_FT_MAX)
655                                 d_type = ocfs2_filetype_table[de->file_type];
656
657                         filldir_ret = filldir(priv, de->name,
658                                               de->name_len,
659                                               *f_pos,
660                                               le64_to_cpu(de->inode),
661                                               d_type);
662                         if (filldir_ret)
663                                 break;
664                         if (version != *f_version)
665                                 goto revalidate;
666                 }
667                 *f_pos += le16_to_cpu(de->rec_len);
668         }
669
670 out:
671         brelse(di_bh);
672
673         return 0;
674 }
675
676 static int ocfs2_dir_foreach_blk_el(struct inode *inode,
677                                     unsigned long *f_version,
678                                     loff_t *f_pos, void *priv,
679                                     filldir_t filldir)
680 {
681         int error = 0;
682         unsigned long offset, blk, last_ra_blk = 0;
683         int i, stored;
684         struct buffer_head * bh, * tmp;
685         struct ocfs2_dir_entry * de;
686         int err;
687         struct super_block * sb = inode->i_sb;
688         unsigned int ra_sectors = 16;
689
690         stored = 0;
691         bh = NULL;
692
693         offset = (*f_pos) & (sb->s_blocksize - 1);
694
695         while (!error && !stored && *f_pos < i_size_read(inode)) {
696                 blk = (*f_pos) >> sb->s_blocksize_bits;
697                 bh = ocfs2_bread(inode, blk, &err, 0);
698                 if (!bh) {
699                         mlog(ML_ERROR,
700                              "directory #%llu contains a hole at offset %lld\n",
701                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
702                              *f_pos);
703                         *f_pos += sb->s_blocksize - offset;
704                         continue;
705                 }
706
707                 /* The idea here is to begin with 8k read-ahead and to stay
708                  * 4k ahead of our current position.
709                  *
710                  * TODO: Use the pagecache for this. We just need to
711                  * make sure it's cluster-safe... */
712                 if (!last_ra_blk
713                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
714                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
715                              i > 0; i--) {
716                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
717                                 if (tmp)
718                                         brelse(tmp);
719                         }
720                         last_ra_blk = blk;
721                         ra_sectors = 8;
722                 }
723
724 revalidate:
725                 /* If the dir block has changed since the last call to
726                  * readdir(2), then we might be pointing to an invalid
727                  * dirent right now.  Scan from the start of the block
728                  * to make sure. */
729                 if (*f_version != inode->i_version) {
730                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
731                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
732                                 /* It's too expensive to do a full
733                                  * dirent test each time round this
734                                  * loop, but we do have to test at
735                                  * least that it is non-zero.  A
736                                  * failure will be detected in the
737                                  * dirent test below. */
738                                 if (le16_to_cpu(de->rec_len) <
739                                     OCFS2_DIR_REC_LEN(1))
740                                         break;
741                                 i += le16_to_cpu(de->rec_len);
742                         }
743                         offset = i;
744                         *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
745                                 | offset;
746                         *f_version = inode->i_version;
747                 }
748
749                 while (!error && *f_pos < i_size_read(inode)
750                        && offset < sb->s_blocksize) {
751                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
752                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
753                                 /* On error, skip the f_pos to the
754                                    next block. */
755                                 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
756                                 brelse(bh);
757                                 goto out;
758                         }
759                         offset += le16_to_cpu(de->rec_len);
760                         if (le64_to_cpu(de->inode)) {
761                                 /* We might block in the next section
762                                  * if the data destination is
763                                  * currently swapped out.  So, use a
764                                  * version stamp to detect whether or
765                                  * not the directory has been modified
766                                  * during the copy operation.
767                                  */
768                                 unsigned long version = *f_version;
769                                 unsigned char d_type = DT_UNKNOWN;
770
771                                 if (de->file_type < OCFS2_FT_MAX)
772                                         d_type = ocfs2_filetype_table[de->file_type];
773                                 error = filldir(priv, de->name,
774                                                 de->name_len,
775                                                 *f_pos,
776                                                 le64_to_cpu(de->inode),
777                                                 d_type);
778                                 if (error)
779                                         break;
780                                 if (version != *f_version)
781                                         goto revalidate;
782                                 stored ++;
783                         }
784                         *f_pos += le16_to_cpu(de->rec_len);
785                 }
786                 offset = 0;
787                 brelse(bh);
788         }
789
790         stored = 0;
791 out:
792         return stored;
793 }
794
795 static int ocfs2_dir_foreach_blk(struct inode *inode, unsigned long *f_version,
796                                  loff_t *f_pos, void *priv, filldir_t filldir)
797 {
798         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
799                 return ocfs2_dir_foreach_blk_id(inode, f_version, f_pos, priv,
800                                                 filldir);
801
802         return ocfs2_dir_foreach_blk_el(inode, f_version, f_pos, priv, filldir);
803 }
804
805 /*
806  * This is intended to be called from inside other kernel functions,
807  * so we fake some arguments.
808  */
809 int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
810                       filldir_t filldir)
811 {
812         int ret = 0;
813         unsigned long version = inode->i_version;
814
815         while (*f_pos < i_size_read(inode)) {
816                 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
817                                             filldir);
818                 if (ret)
819                         break;
820         }
821
822         return 0;
823 }
824
825 /*
826  * ocfs2_readdir()
827  *
828  */
829 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
830 {
831         int error = 0;
832         struct inode *inode = filp->f_path.dentry->d_inode;
833         int lock_level = 0;
834
835         mlog_entry("dirino=%llu\n",
836                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
837
838         error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
839         if (lock_level && error >= 0) {
840                 /* We release EX lock which used to update atime
841                  * and get PR lock again to reduce contention
842                  * on commonly accessed directories. */
843                 ocfs2_meta_unlock(inode, 1);
844                 lock_level = 0;
845                 error = ocfs2_meta_lock(inode, NULL, 0);
846         }
847         if (error < 0) {
848                 if (error != -ENOENT)
849                         mlog_errno(error);
850                 /* we haven't got any yet, so propagate the error. */
851                 goto bail_nolock;
852         }
853
854         error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
855                                       dirent, filldir);
856
857         ocfs2_meta_unlock(inode, lock_level);
858
859 bail_nolock:
860         mlog_exit(error);
861
862         return error;
863 }
864
865 /*
866  * NOTE: this should always be called with parent dir i_mutex taken.
867  */
868 int ocfs2_find_files_on_disk(const char *name,
869                              int namelen,
870                              u64 *blkno,
871                              struct inode *inode,
872                              struct buffer_head **dirent_bh,
873                              struct ocfs2_dir_entry **dirent)
874 {
875         int status = -ENOENT;
876
877         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
878                    namelen, name, blkno, inode, dirent_bh, dirent);
879
880         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
881         if (!*dirent_bh || !*dirent) {
882                 status = -ENOENT;
883                 goto leave;
884         }
885
886         *blkno = le64_to_cpu((*dirent)->inode);
887
888         status = 0;
889 leave:
890         if (status < 0) {
891                 *dirent = NULL;
892                 if (*dirent_bh) {
893                         brelse(*dirent_bh);
894                         *dirent_bh = NULL;
895                 }
896         }
897
898         mlog_exit(status);
899         return status;
900 }
901
902 /*
903  * Convenience function for callers which just want the block number
904  * mapped to a name and don't require the full dirent info, etc.
905  */
906 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
907                                int namelen, u64 *blkno)
908 {
909         int ret;
910         struct buffer_head *bh = NULL;
911         struct ocfs2_dir_entry *dirent = NULL;
912
913         ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &bh, &dirent);
914         brelse(bh);
915
916         return ret;
917 }
918
919 /* Check for a name within a directory.
920  *
921  * Return 0 if the name does not exist
922  * Return -EEXIST if the directory contains the name
923  *
924  * Callers should have i_mutex + a cluster lock on dir
925  */
926 int ocfs2_check_dir_for_entry(struct inode *dir,
927                               const char *name,
928                               int namelen)
929 {
930         int ret;
931         struct buffer_head *dirent_bh = NULL;
932         struct ocfs2_dir_entry *dirent = NULL;
933
934         mlog_entry("dir %llu, name '%.*s'\n",
935                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
936
937         ret = -EEXIST;
938         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
939         if (dirent_bh)
940                 goto bail;
941
942         ret = 0;
943 bail:
944         if (dirent_bh)
945                 brelse(dirent_bh);
946
947         mlog_exit(ret);
948         return ret;
949 }
950
951 struct ocfs2_empty_dir_priv {
952         unsigned seen_dot;
953         unsigned seen_dot_dot;
954         unsigned seen_other;
955 };
956 static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
957                                    loff_t pos, u64 ino, unsigned type)
958 {
959         struct ocfs2_empty_dir_priv *p = priv;
960
961         /*
962          * Check the positions of "." and ".." records to be sure
963          * they're in the correct place.
964          */
965         if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
966                 p->seen_dot = 1;
967                 return 0;
968         }
969
970         if (name_len == 2 && !strncmp("..", name, 2) &&
971             pos == OCFS2_DIR_REC_LEN(1)) {
972                 p->seen_dot_dot = 1;
973                 return 0;
974         }
975
976         p->seen_other = 1;
977         return 1;
978 }
979 /*
980  * routine to check that the specified directory is empty (for rmdir)
981  *
982  * Returns 1 if dir is empty, zero otherwise.
983  */
984 int ocfs2_empty_dir(struct inode *inode)
985 {
986         int ret;
987         loff_t start = 0;
988         struct ocfs2_empty_dir_priv priv;
989
990         memset(&priv, 0, sizeof(priv));
991
992         ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
993         if (ret)
994                 mlog_errno(ret);
995
996         if (!priv.seen_dot || !priv.seen_dot_dot) {
997                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
998                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
999                 /*
1000                  * XXX: Is it really safe to allow an unlink to continue?
1001                  */
1002                 return 1;
1003         }
1004
1005         return !priv.seen_other;
1006 }
1007
1008 static void ocfs2_fill_initial_dirents(struct inode *inode,
1009                                        struct inode *parent,
1010                                        char *start, unsigned int size)
1011 {
1012         struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
1013
1014         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
1015         de->name_len = 1;
1016         de->rec_len =
1017                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1018         strcpy(de->name, ".");
1019         ocfs2_set_de_type(de, S_IFDIR);
1020
1021         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
1022         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
1023         de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
1024         de->name_len = 2;
1025         strcpy(de->name, "..");
1026         ocfs2_set_de_type(de, S_IFDIR);
1027 }
1028
1029 /*
1030  * This works together with code in ocfs2_mknod_locked() which sets
1031  * the inline-data flag and initializes the inline-data section.
1032  */
1033 static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
1034                                  handle_t *handle,
1035                                  struct inode *parent,
1036                                  struct inode *inode,
1037                                  struct buffer_head *di_bh)
1038 {
1039         int ret;
1040         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1041         struct ocfs2_inline_data *data = &di->id2.i_data;
1042         unsigned int size = le16_to_cpu(data->id_count);
1043
1044         ret = ocfs2_journal_access(handle, inode, di_bh,
1045                                    OCFS2_JOURNAL_ACCESS_WRITE);
1046         if (ret) {
1047                 mlog_errno(ret);
1048                 goto out;
1049         }
1050
1051         ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
1052
1053         ocfs2_journal_dirty(handle, di_bh);
1054         if (ret) {
1055                 mlog_errno(ret);
1056                 goto out;
1057         }
1058
1059         i_size_write(inode, size);
1060         inode->i_nlink = 2;
1061         inode->i_blocks = ocfs2_inode_sector_count(inode);
1062
1063         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1064         if (ret < 0)
1065                 mlog_errno(ret);
1066
1067 out:
1068         return ret;
1069 }
1070
1071 static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
1072                                  handle_t *handle,
1073                                  struct inode *parent,
1074                                  struct inode *inode,
1075                                  struct buffer_head *fe_bh,
1076                                  struct ocfs2_alloc_context *data_ac)
1077 {
1078         int status;
1079         struct buffer_head *new_bh = NULL;
1080
1081         mlog_entry_void();
1082
1083         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
1084                                      data_ac, NULL, &new_bh);
1085         if (status < 0) {
1086                 mlog_errno(status);
1087                 goto bail;
1088         }
1089
1090         ocfs2_set_new_buffer_uptodate(inode, new_bh);
1091
1092         status = ocfs2_journal_access(handle, inode, new_bh,
1093                                       OCFS2_JOURNAL_ACCESS_CREATE);
1094         if (status < 0) {
1095                 mlog_errno(status);
1096                 goto bail;
1097         }
1098         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
1099
1100         ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data,
1101                                    osb->sb->s_blocksize);
1102
1103         status = ocfs2_journal_dirty(handle, new_bh);
1104         if (status < 0) {
1105                 mlog_errno(status);
1106                 goto bail;
1107         }
1108
1109         i_size_write(inode, inode->i_sb->s_blocksize);
1110         inode->i_nlink = 2;
1111         inode->i_blocks = ocfs2_inode_sector_count(inode);
1112         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
1113         if (status < 0) {
1114                 mlog_errno(status);
1115                 goto bail;
1116         }
1117
1118         status = 0;
1119 bail:
1120         if (new_bh)
1121                 brelse(new_bh);
1122
1123         mlog_exit(status);
1124         return status;
1125 }
1126
1127 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
1128                        handle_t *handle,
1129                        struct inode *parent,
1130                        struct inode *inode,
1131                        struct buffer_head *fe_bh,
1132                        struct ocfs2_alloc_context *data_ac)
1133 {
1134         BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
1135
1136         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1137                 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
1138
1139         return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
1140                                      data_ac);
1141 }
1142
1143 static void ocfs2_expand_last_dirent(char *start, unsigned int old_size,
1144                                      unsigned int new_size)
1145 {
1146         struct ocfs2_dir_entry *de;
1147         struct ocfs2_dir_entry *prev_de;
1148         char *de_buf, *limit;
1149         unsigned int bytes = new_size - old_size;
1150
1151         limit = start + old_size;
1152         de_buf = start;
1153         de = (struct ocfs2_dir_entry *)de_buf;
1154         do {
1155                 prev_de = de;
1156                 de_buf += le16_to_cpu(de->rec_len);
1157                 de = (struct ocfs2_dir_entry *)de_buf;
1158         } while (de_buf < limit);
1159
1160         le16_add_cpu(&prev_de->rec_len, bytes);
1161 }
1162
1163 /*
1164  * We allocate enough clusters to fulfill "blocks_wanted", but set
1165  * i_size to exactly one block. Ocfs2_extend_dir() will handle the
1166  * rest automatically for us.
1167  *
1168  * *first_block_bh is a pointer to the 1st data block allocated to the
1169  *  directory.
1170  */
1171 static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
1172                                    unsigned int blocks_wanted,
1173                                    struct buffer_head **first_block_bh)
1174 {
1175         int ret, credits = OCFS2_INLINE_TO_EXTENTS_CREDITS;
1176         u32 alloc, bit_off, len;
1177         struct super_block *sb = dir->i_sb;
1178         u64 blkno, bytes = blocks_wanted << sb->s_blocksize_bits;
1179         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
1180         struct ocfs2_inode_info *oi = OCFS2_I(dir);
1181         struct ocfs2_alloc_context *data_ac;
1182         struct buffer_head *dirdata_bh = NULL;
1183         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1184         handle_t *handle;
1185
1186         alloc = ocfs2_clusters_for_bytes(sb, bytes);
1187
1188         /*
1189          * We should never need more than 2 clusters for this -
1190          * maximum dirent size is far less than one block. In fact,
1191          * the only time we'd need more than one cluster is if
1192          * blocksize == clustersize and the dirent won't fit in the
1193          * extra space that the expansion to a single block gives. As
1194          * of today, that only happens on 4k/4k file systems.
1195          */
1196         BUG_ON(alloc > 2);
1197
1198         ret = ocfs2_reserve_clusters(osb, alloc, &data_ac);
1199         if (ret) {
1200                 mlog_errno(ret);
1201                 goto out;
1202         }
1203
1204         down_write(&oi->ip_alloc_sem);
1205
1206         /*
1207          * Prepare for worst case allocation scenario of two seperate
1208          * extents.
1209          */
1210         if (alloc == 2)
1211                 credits += OCFS2_SUBALLOC_ALLOC;
1212
1213         handle = ocfs2_start_trans(osb, credits);
1214         if (IS_ERR(handle)) {
1215                 ret = PTR_ERR(handle);
1216                 mlog_errno(ret);
1217                 goto out_sem;
1218         }
1219
1220         /*
1221          * Try to claim as many clusters as the bitmap can give though
1222          * if we only get one now, that's enough to continue. The rest
1223          * will be claimed after the conversion to extents.
1224          */
1225         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
1226         if (ret) {
1227                 mlog_errno(ret);
1228                 goto out_commit;
1229         }
1230
1231         /*
1232          * Operations are carefully ordered so that we set up the new
1233          * data block first. The conversion from inline data to
1234          * extents follows.
1235          */
1236         blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
1237         dirdata_bh = sb_getblk(sb, blkno);
1238         if (!dirdata_bh) {
1239                 ret = -EIO;
1240                 mlog_errno(ret);
1241                 goto out_commit;
1242         }
1243
1244         ocfs2_set_new_buffer_uptodate(dir, dirdata_bh);
1245
1246         ret = ocfs2_journal_access(handle, dir, dirdata_bh,
1247                                    OCFS2_JOURNAL_ACCESS_CREATE);
1248         if (ret) {
1249                 mlog_errno(ret);
1250                 goto out_commit;
1251         }
1252
1253         memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
1254         memset(dirdata_bh->b_data + i_size_read(dir), 0,
1255                sb->s_blocksize - i_size_read(dir));
1256         ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir),
1257                                  sb->s_blocksize);
1258
1259         ret = ocfs2_journal_dirty(handle, dirdata_bh);
1260         if (ret) {
1261                 mlog_errno(ret);
1262                 goto out_commit;
1263         }
1264
1265         /*
1266          * Set extent, i_size, etc on the directory. After this, the
1267          * inode should contain the same exact dirents as before and
1268          * be fully accessible from system calls.
1269          *
1270          * We let the later dirent insert modify c/mtime - to the user
1271          * the data hasn't changed.
1272          */
1273         ret = ocfs2_journal_access(handle, dir, di_bh,
1274                                    OCFS2_JOURNAL_ACCESS_CREATE);
1275         if (ret) {
1276                 mlog_errno(ret);
1277                 goto out_commit;
1278         }
1279
1280         spin_lock(&oi->ip_lock);
1281         oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
1282         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1283         spin_unlock(&oi->ip_lock);
1284
1285         ocfs2_dinode_new_extent_list(dir, di);
1286
1287         i_size_write(dir, sb->s_blocksize);
1288         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1289
1290         di->i_size = cpu_to_le64(sb->s_blocksize);
1291         di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
1292         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
1293         dir->i_blocks = ocfs2_inode_sector_count(dir);
1294
1295         /*
1296          * This should never fail as our extent list is empty and all
1297          * related blocks have been journaled already.
1298          */
1299         ret = ocfs2_insert_extent(osb, handle, dir, di_bh, 0, blkno, len, 0,
1300                                   NULL);
1301         if (ret) {
1302                 mlog_errno(ret);
1303                 goto out;
1304         }
1305
1306         ret = ocfs2_journal_dirty(handle, di_bh);
1307         if (ret) {
1308                 mlog_errno(ret);
1309                 goto out_commit;
1310         }
1311
1312         /*
1313          * We asked for two clusters, but only got one in the 1st
1314          * pass. Claim the 2nd cluster as a separate extent.
1315          */
1316         if (alloc > len) {
1317                 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
1318                                            &len);
1319                 if (ret) {
1320                         mlog_errno(ret);
1321                         goto out_commit;
1322                 }
1323                 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
1324
1325                 ret = ocfs2_insert_extent(osb, handle, dir, di_bh, 1, blkno,
1326                                           len, 0, NULL);
1327                 if (ret) {
1328                         mlog_errno(ret);
1329                         goto out;
1330                 }
1331         }
1332
1333         *first_block_bh = dirdata_bh;
1334         dirdata_bh = NULL;
1335
1336 out_commit:
1337         ocfs2_commit_trans(osb, handle);
1338
1339 out_sem:
1340         up_write(&oi->ip_alloc_sem);
1341
1342 out:
1343         if (data_ac)
1344                 ocfs2_free_alloc_context(data_ac);
1345
1346         brelse(dirdata_bh);
1347
1348         return ret;
1349 }
1350
1351 /* returns a bh of the 1st new block in the allocation. */
1352 static int ocfs2_do_extend_dir(struct super_block *sb,
1353                                handle_t *handle,
1354                                struct inode *dir,
1355                                struct buffer_head *parent_fe_bh,
1356                                struct ocfs2_alloc_context *data_ac,
1357                                struct ocfs2_alloc_context *meta_ac,
1358                                struct buffer_head **new_bh)
1359 {
1360         int status;
1361         int extend;
1362         u64 p_blkno, v_blkno;
1363
1364         spin_lock(&OCFS2_I(dir)->ip_lock);
1365         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
1366         spin_unlock(&OCFS2_I(dir)->ip_lock);
1367
1368         if (extend) {
1369                 u32 offset = OCFS2_I(dir)->ip_clusters;
1370
1371                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
1372                                                     1, 0, parent_fe_bh, handle,
1373                                                     data_ac, meta_ac, NULL);
1374                 BUG_ON(status == -EAGAIN);
1375                 if (status < 0) {
1376                         mlog_errno(status);
1377                         goto bail;
1378                 }
1379         }
1380
1381         v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
1382         status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
1383         if (status < 0) {
1384                 mlog_errno(status);
1385                 goto bail;
1386         }
1387
1388         *new_bh = sb_getblk(sb, p_blkno);
1389         if (!*new_bh) {
1390                 status = -EIO;
1391                 mlog_errno(status);
1392                 goto bail;
1393         }
1394         status = 0;
1395 bail:
1396         mlog_exit(status);
1397         return status;
1398 }
1399
1400 /*
1401  * Assumes you already have a cluster lock on the directory.
1402  *
1403  * 'blocks_wanted' is only used if we have an inline directory which
1404  * is to be turned into an extent based one. The size of the dirent to
1405  * insert might be larger than the space gained by growing to just one
1406  * block, so we may have to grow the inode by two blocks in that case.
1407  */
1408 static int ocfs2_extend_dir(struct ocfs2_super *osb,
1409                             struct inode *dir,
1410                             struct buffer_head *parent_fe_bh,
1411                             unsigned int blocks_wanted,
1412                             struct buffer_head **new_de_bh)
1413 {
1414         int status = 0;
1415         int credits, num_free_extents, drop_alloc_sem = 0;
1416         loff_t dir_i_size;
1417         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1418         struct ocfs2_alloc_context *data_ac = NULL;
1419         struct ocfs2_alloc_context *meta_ac = NULL;
1420         handle_t *handle = NULL;
1421         struct buffer_head *new_bh = NULL;
1422         struct ocfs2_dir_entry * de;
1423         struct super_block *sb = osb->sb;
1424
1425         mlog_entry_void();
1426
1427         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1428                 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
1429                                                  blocks_wanted, &new_bh);
1430                 if (status) {
1431                         mlog_errno(status);
1432                         goto bail;
1433                 }
1434
1435                 if (blocks_wanted == 1) {
1436                         /*
1437                          * If the new dirent will fit inside the space
1438                          * created by pushing out to one block, then
1439                          * we can complete the operation
1440                          * here. Otherwise we have to expand i_size
1441                          * and format the 2nd block below.
1442                          */
1443                         BUG_ON(new_bh == NULL);
1444                         goto bail_bh;
1445                 }
1446
1447                 /*
1448                  * Get rid of 'new_bh' - we want to format the 2nd
1449                  * data block and return that instead.
1450                  */
1451                 brelse(new_bh);
1452                 new_bh = NULL;
1453
1454                 dir_i_size = i_size_read(dir);
1455                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
1456                 goto do_extend;
1457         }
1458
1459         dir_i_size = i_size_read(dir);
1460         mlog(0, "extending dir %llu (i_size = %lld)\n",
1461              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
1462
1463         /* dir->i_size is always block aligned. */
1464         spin_lock(&OCFS2_I(dir)->ip_lock);
1465         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
1466                 spin_unlock(&OCFS2_I(dir)->ip_lock);
1467                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
1468                 if (num_free_extents < 0) {
1469                         status = num_free_extents;
1470                         mlog_errno(status);
1471                         goto bail;
1472                 }
1473
1474                 if (!num_free_extents) {
1475                         status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
1476                         if (status < 0) {
1477                                 if (status != -ENOSPC)
1478                                         mlog_errno(status);
1479                                 goto bail;
1480                         }
1481                 }
1482
1483                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1484                 if (status < 0) {
1485                         if (status != -ENOSPC)
1486                                 mlog_errno(status);
1487                         goto bail;
1488                 }
1489
1490                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
1491         } else {
1492                 spin_unlock(&OCFS2_I(dir)->ip_lock);
1493                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
1494         }
1495
1496 do_extend:
1497         down_write(&OCFS2_I(dir)->ip_alloc_sem);
1498         drop_alloc_sem = 1;
1499
1500         handle = ocfs2_start_trans(osb, credits);
1501         if (IS_ERR(handle)) {
1502                 status = PTR_ERR(handle);
1503                 handle = NULL;
1504                 mlog_errno(status);
1505                 goto bail;
1506         }
1507
1508         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
1509                                      data_ac, meta_ac, &new_bh);
1510         if (status < 0) {
1511                 mlog_errno(status);
1512                 goto bail;
1513         }
1514
1515         ocfs2_set_new_buffer_uptodate(dir, new_bh);
1516
1517         status = ocfs2_journal_access(handle, dir, new_bh,
1518                                       OCFS2_JOURNAL_ACCESS_CREATE);
1519         if (status < 0) {
1520                 mlog_errno(status);
1521                 goto bail;
1522         }
1523         memset(new_bh->b_data, 0, sb->s_blocksize);
1524         de = (struct ocfs2_dir_entry *) new_bh->b_data;
1525         de->inode = 0;
1526         de->rec_len = cpu_to_le16(sb->s_blocksize);
1527         status = ocfs2_journal_dirty(handle, new_bh);
1528         if (status < 0) {
1529                 mlog_errno(status);
1530                 goto bail;
1531         }
1532
1533         dir_i_size += dir->i_sb->s_blocksize;
1534         i_size_write(dir, dir_i_size);
1535         dir->i_blocks = ocfs2_inode_sector_count(dir);
1536         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1537         if (status < 0) {
1538                 mlog_errno(status);
1539                 goto bail;
1540         }
1541
1542 bail_bh:
1543         *new_de_bh = new_bh;
1544         get_bh(*new_de_bh);
1545 bail:
1546         if (drop_alloc_sem)
1547                 up_write(&OCFS2_I(dir)->ip_alloc_sem);
1548         if (handle)
1549                 ocfs2_commit_trans(osb, handle);
1550
1551         if (data_ac)
1552                 ocfs2_free_alloc_context(data_ac);
1553         if (meta_ac)
1554                 ocfs2_free_alloc_context(meta_ac);
1555
1556         if (new_bh)
1557                 brelse(new_bh);
1558
1559         mlog_exit(status);
1560         return status;
1561 }
1562
1563 static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
1564                                    const char *name, int namelen,
1565                                    struct buffer_head **ret_de_bh,
1566                                    unsigned int *blocks_wanted)
1567 {
1568         int ret;
1569         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1570         struct ocfs2_dir_entry *de, *last_de = NULL;
1571         char *de_buf, *limit;
1572         unsigned long offset = 0;
1573         unsigned int rec_len, new_rec_len;
1574
1575         de_buf = di->id2.i_data.id_data;
1576         limit = de_buf + i_size_read(dir);
1577         rec_len = OCFS2_DIR_REC_LEN(namelen);
1578
1579         while (de_buf < limit) {
1580                 de = (struct ocfs2_dir_entry *)de_buf;
1581
1582                 if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
1583                         ret = -ENOENT;
1584                         goto out;
1585                 }
1586                 if (ocfs2_match(namelen, name, de)) {
1587                         ret = -EEXIST;
1588                         goto out;
1589                 }
1590                 if (ocfs2_dirent_would_fit(de, rec_len)) {
1591                         /* Ok, we found a spot. Return this bh and let
1592                          * the caller actually fill it in. */
1593                         *ret_de_bh = di_bh;
1594                         get_bh(*ret_de_bh);
1595                         ret = 0;
1596                         goto out;
1597                 }
1598
1599                 last_de = de;
1600                 de_buf += le16_to_cpu(de->rec_len);
1601                 offset += le16_to_cpu(de->rec_len);
1602         }
1603
1604         /*
1605          * We're going to require expansion of the directory - figure
1606          * out how many blocks we'll need so that a place for the
1607          * dirent can be found.
1608          */
1609         *blocks_wanted = 1;
1610         new_rec_len = le16_to_cpu(last_de->rec_len) + (dir->i_sb->s_blocksize - i_size_read(dir));
1611         if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
1612                 *blocks_wanted = 2;
1613
1614         ret = -ENOSPC;
1615 out:
1616         return ret;
1617 }
1618
1619 static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
1620                                    int namelen, struct buffer_head **ret_de_bh)
1621 {
1622         unsigned long offset;
1623         struct buffer_head *bh = NULL;
1624         unsigned short rec_len;
1625         struct ocfs2_dir_entry *de;
1626         struct super_block *sb = dir->i_sb;
1627         int status;
1628
1629         bh = ocfs2_bread(dir, 0, &status, 0);
1630         if (!bh) {
1631                 mlog_errno(status);
1632                 goto bail;
1633         }
1634
1635         rec_len = OCFS2_DIR_REC_LEN(namelen);
1636         offset = 0;
1637         de = (struct ocfs2_dir_entry *) bh->b_data;
1638         while (1) {
1639                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
1640                         brelse(bh);
1641                         bh = NULL;
1642
1643                         if (i_size_read(dir) <= offset) {
1644                                 /*
1645                                  * Caller will have to expand this
1646                                  * directory.
1647                                  */
1648                                 status = -ENOSPC;
1649                                 goto bail;
1650                         }
1651                         bh = ocfs2_bread(dir,
1652                                          offset >> sb->s_blocksize_bits,
1653                                          &status,
1654                                          0);
1655                         if (!bh) {
1656                                 mlog_errno(status);
1657                                 goto bail;
1658                         }
1659                         /* move to next block */
1660                         de = (struct ocfs2_dir_entry *) bh->b_data;
1661                 }
1662                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1663                         status = -ENOENT;
1664                         goto bail;
1665                 }
1666                 if (ocfs2_match(namelen, name, de)) {
1667                         status = -EEXIST;
1668                         goto bail;
1669                 }
1670                 if (ocfs2_dirent_would_fit(de, rec_len)) {
1671                         /* Ok, we found a spot. Return this bh and let
1672                          * the caller actually fill it in. */
1673                         *ret_de_bh = bh;
1674                         get_bh(*ret_de_bh);
1675                         status = 0;
1676                         goto bail;
1677                 }
1678                 offset += le16_to_cpu(de->rec_len);
1679                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
1680         }
1681
1682         status = 0;
1683 bail:
1684         if (bh)
1685                 brelse(bh);
1686
1687         mlog_exit(status);
1688         return status;
1689 }
1690
1691 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
1692                                  struct inode *dir,
1693                                  struct buffer_head *parent_fe_bh,
1694                                  const char *name,
1695                                  int namelen,
1696                                  struct buffer_head **ret_de_bh)
1697 {
1698         int ret;
1699         unsigned int blocks_wanted = 1;
1700         struct buffer_head *bh = NULL;
1701
1702         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
1703              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
1704
1705         *ret_de_bh = NULL;
1706
1707         if (!namelen) {
1708                 ret = -EINVAL;
1709                 mlog_errno(ret);
1710                 goto out;
1711         }
1712
1713         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1714                 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
1715                                               namelen, &bh, &blocks_wanted);
1716         } else
1717                 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
1718
1719         if (ret && ret != -ENOSPC) {
1720                 mlog_errno(ret);
1721                 goto out;
1722         }
1723
1724         if (ret == -ENOSPC) {
1725                 /*
1726                  * We have to expand the directory to add this name.
1727                  */
1728                 BUG_ON(bh);
1729
1730                 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
1731                                        &bh);
1732                 if (ret) {
1733                         if (ret != -ENOSPC)
1734                                 mlog_errno(ret);
1735                         goto out;
1736                 }
1737
1738                 BUG_ON(!bh);
1739         }
1740
1741         *ret_de_bh = bh;
1742         bh = NULL;
1743 out:
1744         if (bh)
1745                 brelse(bh);
1746         return ret;
1747 }