Merge branch 'pnfs-submit' of git://git.open-osd.org/linux-open-osd
[linux-2.6.git] / fs / hfsplus / extents.c
1 /*
2  *  linux/fs/hfsplus/extents.c
3  *
4  * Copyright (C) 2001
5  * Brad Boyer (flar@allandria.com)
6  * (C) 2003 Ardis Technologies <roman@ardistech.com>
7  *
8  * Handling of Extents both in catalog and extents overflow trees
9  */
10
11 #include <linux/errno.h>
12 #include <linux/fs.h>
13 #include <linux/pagemap.h>
14
15 #include "hfsplus_fs.h"
16 #include "hfsplus_raw.h"
17
18 /* Compare two extents keys, returns 0 on same, pos/neg for difference */
19 int hfsplus_ext_cmp_key(const hfsplus_btree_key *k1,
20                         const hfsplus_btree_key *k2)
21 {
22         __be32 k1id, k2id;
23         __be32 k1s, k2s;
24
25         k1id = k1->ext.cnid;
26         k2id = k2->ext.cnid;
27         if (k1id != k2id)
28                 return be32_to_cpu(k1id) < be32_to_cpu(k2id) ? -1 : 1;
29
30         if (k1->ext.fork_type != k2->ext.fork_type)
31                 return k1->ext.fork_type < k2->ext.fork_type ? -1 : 1;
32
33         k1s = k1->ext.start_block;
34         k2s = k2->ext.start_block;
35         if (k1s == k2s)
36                 return 0;
37         return be32_to_cpu(k1s) < be32_to_cpu(k2s) ? -1 : 1;
38 }
39
40 static void hfsplus_ext_build_key(hfsplus_btree_key *key, u32 cnid,
41                                   u32 block, u8 type)
42 {
43         key->key_len = cpu_to_be16(HFSPLUS_EXT_KEYLEN - 2);
44         key->ext.cnid = cpu_to_be32(cnid);
45         key->ext.start_block = cpu_to_be32(block);
46         key->ext.fork_type = type;
47         key->ext.pad = 0;
48 }
49
50 static u32 hfsplus_ext_find_block(struct hfsplus_extent *ext, u32 off)
51 {
52         int i;
53         u32 count;
54
55         for (i = 0; i < 8; ext++, i++) {
56                 count = be32_to_cpu(ext->block_count);
57                 if (off < count)
58                         return be32_to_cpu(ext->start_block) + off;
59                 off -= count;
60         }
61         /* panic? */
62         return 0;
63 }
64
65 static int hfsplus_ext_block_count(struct hfsplus_extent *ext)
66 {
67         int i;
68         u32 count = 0;
69
70         for (i = 0; i < 8; ext++, i++)
71                 count += be32_to_cpu(ext->block_count);
72         return count;
73 }
74
75 static u32 hfsplus_ext_lastblock(struct hfsplus_extent *ext)
76 {
77         int i;
78
79         ext += 7;
80         for (i = 0; i < 7; ext--, i++)
81                 if (ext->block_count)
82                         break;
83         return be32_to_cpu(ext->start_block) + be32_to_cpu(ext->block_count);
84 }
85
86 static void __hfsplus_ext_write_extent(struct inode *inode,
87                 struct hfs_find_data *fd)
88 {
89         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
90         int res;
91
92         WARN_ON(!mutex_is_locked(&hip->extents_lock));
93
94         hfsplus_ext_build_key(fd->search_key, inode->i_ino, hip->cached_start,
95                               HFSPLUS_IS_RSRC(inode) ?
96                                 HFSPLUS_TYPE_RSRC : HFSPLUS_TYPE_DATA);
97
98         res = hfs_brec_find(fd);
99         if (hip->extent_state & HFSPLUS_EXT_NEW) {
100                 if (res != -ENOENT)
101                         return;
102                 hfs_brec_insert(fd, hip->cached_extents,
103                                 sizeof(hfsplus_extent_rec));
104                 hip->extent_state &= ~(HFSPLUS_EXT_DIRTY | HFSPLUS_EXT_NEW);
105         } else {
106                 if (res)
107                         return;
108                 hfs_bnode_write(fd->bnode, hip->cached_extents,
109                                 fd->entryoffset, fd->entrylength);
110                 hip->extent_state &= ~HFSPLUS_EXT_DIRTY;
111         }
112
113         /*
114          * We can't just use hfsplus_mark_inode_dirty here, because we
115          * also get called from hfsplus_write_inode, which should not
116          * redirty the inode.  Instead the callers have to be careful
117          * to explicily mark the inode dirty, too.
118          */
119         set_bit(HFSPLUS_I_EXT_DIRTY, &hip->flags);
120 }
121
122 static void hfsplus_ext_write_extent_locked(struct inode *inode)
123 {
124         if (HFSPLUS_I(inode)->extent_state & HFSPLUS_EXT_DIRTY) {
125                 struct hfs_find_data fd;
126
127                 hfs_find_init(HFSPLUS_SB(inode->i_sb)->ext_tree, &fd);
128                 __hfsplus_ext_write_extent(inode, &fd);
129                 hfs_find_exit(&fd);
130         }
131 }
132
133 void hfsplus_ext_write_extent(struct inode *inode)
134 {
135         mutex_lock(&HFSPLUS_I(inode)->extents_lock);
136         hfsplus_ext_write_extent_locked(inode);
137         mutex_unlock(&HFSPLUS_I(inode)->extents_lock);
138 }
139
140 static inline int __hfsplus_ext_read_extent(struct hfs_find_data *fd,
141                                             struct hfsplus_extent *extent,
142                                             u32 cnid, u32 block, u8 type)
143 {
144         int res;
145
146         hfsplus_ext_build_key(fd->search_key, cnid, block, type);
147         fd->key->ext.cnid = 0;
148         res = hfs_brec_find(fd);
149         if (res && res != -ENOENT)
150                 return res;
151         if (fd->key->ext.cnid != fd->search_key->ext.cnid ||
152             fd->key->ext.fork_type != fd->search_key->ext.fork_type)
153                 return -ENOENT;
154         if (fd->entrylength != sizeof(hfsplus_extent_rec))
155                 return -EIO;
156         hfs_bnode_read(fd->bnode, extent, fd->entryoffset,
157                 sizeof(hfsplus_extent_rec));
158         return 0;
159 }
160
161 static inline int __hfsplus_ext_cache_extent(struct hfs_find_data *fd,
162                 struct inode *inode, u32 block)
163 {
164         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
165         int res;
166
167         WARN_ON(!mutex_is_locked(&hip->extents_lock));
168
169         if (hip->extent_state & HFSPLUS_EXT_DIRTY)
170                 __hfsplus_ext_write_extent(inode, fd);
171
172         res = __hfsplus_ext_read_extent(fd, hip->cached_extents, inode->i_ino,
173                                         block, HFSPLUS_IS_RSRC(inode) ?
174                                                 HFSPLUS_TYPE_RSRC :
175                                                 HFSPLUS_TYPE_DATA);
176         if (!res) {
177                 hip->cached_start = be32_to_cpu(fd->key->ext.start_block);
178                 hip->cached_blocks =
179                         hfsplus_ext_block_count(hip->cached_extents);
180         } else {
181                 hip->cached_start = hip->cached_blocks = 0;
182                 hip->extent_state &= ~(HFSPLUS_EXT_DIRTY | HFSPLUS_EXT_NEW);
183         }
184         return res;
185 }
186
187 static int hfsplus_ext_read_extent(struct inode *inode, u32 block)
188 {
189         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
190         struct hfs_find_data fd;
191         int res;
192
193         if (block >= hip->cached_start &&
194             block < hip->cached_start + hip->cached_blocks)
195                 return 0;
196
197         hfs_find_init(HFSPLUS_SB(inode->i_sb)->ext_tree, &fd);
198         res = __hfsplus_ext_cache_extent(&fd, inode, block);
199         hfs_find_exit(&fd);
200         return res;
201 }
202
203 /* Get a block at iblock for inode, possibly allocating if create */
204 int hfsplus_get_block(struct inode *inode, sector_t iblock,
205                       struct buffer_head *bh_result, int create)
206 {
207         struct super_block *sb = inode->i_sb;
208         struct hfsplus_sb_info *sbi = HFSPLUS_SB(sb);
209         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
210         int res = -EIO;
211         u32 ablock, dblock, mask;
212         int was_dirty = 0;
213         int shift;
214
215         /* Convert inode block to disk allocation block */
216         shift = sbi->alloc_blksz_shift - sb->s_blocksize_bits;
217         ablock = iblock >> sbi->fs_shift;
218
219         if (iblock >= hip->fs_blocks) {
220                 if (iblock > hip->fs_blocks || !create)
221                         return -EIO;
222                 if (ablock >= hip->alloc_blocks) {
223                         res = hfsplus_file_extend(inode);
224                         if (res)
225                                 return res;
226                 }
227         } else
228                 create = 0;
229
230         if (ablock < hip->first_blocks) {
231                 dblock = hfsplus_ext_find_block(hip->first_extents, ablock);
232                 goto done;
233         }
234
235         if (inode->i_ino == HFSPLUS_EXT_CNID)
236                 return -EIO;
237
238         mutex_lock(&hip->extents_lock);
239
240         /*
241          * hfsplus_ext_read_extent will write out a cached extent into
242          * the extents btree.  In that case we may have to mark the inode
243          * dirty even for a pure read of an extent here.
244          */
245         was_dirty = (hip->extent_state & HFSPLUS_EXT_DIRTY);
246         res = hfsplus_ext_read_extent(inode, ablock);
247         if (res) {
248                 mutex_unlock(&hip->extents_lock);
249                 return -EIO;
250         }
251         dblock = hfsplus_ext_find_block(hip->cached_extents,
252                                         ablock - hip->cached_start);
253         mutex_unlock(&hip->extents_lock);
254
255 done:
256         dprint(DBG_EXTENT, "get_block(%lu): %llu - %u\n",
257                 inode->i_ino, (long long)iblock, dblock);
258         mask = (1 << sbi->fs_shift) - 1;
259         map_bh(bh_result, sb,
260                 (dblock << sbi->fs_shift) + sbi->blockoffset +
261                         (iblock & mask));
262         if (create) {
263                 set_buffer_new(bh_result);
264                 hip->phys_size += sb->s_blocksize;
265                 hip->fs_blocks++;
266                 inode_add_bytes(inode, sb->s_blocksize);
267         }
268         if (create || was_dirty)
269                 mark_inode_dirty(inode);
270         return 0;
271 }
272
273 static void hfsplus_dump_extent(struct hfsplus_extent *extent)
274 {
275         int i;
276
277         dprint(DBG_EXTENT, "   ");
278         for (i = 0; i < 8; i++)
279                 dprint(DBG_EXTENT, " %u:%u", be32_to_cpu(extent[i].start_block),
280                                  be32_to_cpu(extent[i].block_count));
281         dprint(DBG_EXTENT, "\n");
282 }
283
284 static int hfsplus_add_extent(struct hfsplus_extent *extent, u32 offset,
285                               u32 alloc_block, u32 block_count)
286 {
287         u32 count, start;
288         int i;
289
290         hfsplus_dump_extent(extent);
291         for (i = 0; i < 8; extent++, i++) {
292                 count = be32_to_cpu(extent->block_count);
293                 if (offset == count) {
294                         start = be32_to_cpu(extent->start_block);
295                         if (alloc_block != start + count) {
296                                 if (++i >= 8)
297                                         return -ENOSPC;
298                                 extent++;
299                                 extent->start_block = cpu_to_be32(alloc_block);
300                         } else
301                                 block_count += count;
302                         extent->block_count = cpu_to_be32(block_count);
303                         return 0;
304                 } else if (offset < count)
305                         break;
306                 offset -= count;
307         }
308         /* panic? */
309         return -EIO;
310 }
311
312 static int hfsplus_free_extents(struct super_block *sb,
313                                 struct hfsplus_extent *extent,
314                                 u32 offset, u32 block_nr)
315 {
316         u32 count, start;
317         int i;
318
319         hfsplus_dump_extent(extent);
320         for (i = 0; i < 8; extent++, i++) {
321                 count = be32_to_cpu(extent->block_count);
322                 if (offset == count)
323                         goto found;
324                 else if (offset < count)
325                         break;
326                 offset -= count;
327         }
328         /* panic? */
329         return -EIO;
330 found:
331         for (;;) {
332                 start = be32_to_cpu(extent->start_block);
333                 if (count <= block_nr) {
334                         hfsplus_block_free(sb, start, count);
335                         extent->block_count = 0;
336                         extent->start_block = 0;
337                         block_nr -= count;
338                 } else {
339                         count -= block_nr;
340                         hfsplus_block_free(sb, start + count, block_nr);
341                         extent->block_count = cpu_to_be32(count);
342                         block_nr = 0;
343                 }
344                 if (!block_nr || !i)
345                         return 0;
346                 i--;
347                 extent--;
348                 count = be32_to_cpu(extent->block_count);
349         }
350 }
351
352 int hfsplus_free_fork(struct super_block *sb, u32 cnid,
353                 struct hfsplus_fork_raw *fork, int type)
354 {
355         struct hfs_find_data fd;
356         hfsplus_extent_rec ext_entry;
357         u32 total_blocks, blocks, start;
358         int res, i;
359
360         total_blocks = be32_to_cpu(fork->total_blocks);
361         if (!total_blocks)
362                 return 0;
363
364         blocks = 0;
365         for (i = 0; i < 8; i++)
366                 blocks += be32_to_cpu(fork->extents[i].block_count);
367
368         res = hfsplus_free_extents(sb, fork->extents, blocks, blocks);
369         if (res)
370                 return res;
371         if (total_blocks == blocks)
372                 return 0;
373
374         hfs_find_init(HFSPLUS_SB(sb)->ext_tree, &fd);
375         do {
376                 res = __hfsplus_ext_read_extent(&fd, ext_entry, cnid,
377                                                 total_blocks, type);
378                 if (res)
379                         break;
380                 start = be32_to_cpu(fd.key->ext.start_block);
381                 hfsplus_free_extents(sb, ext_entry,
382                                      total_blocks - start,
383                                      total_blocks);
384                 hfs_brec_remove(&fd);
385                 total_blocks = start;
386         } while (total_blocks > blocks);
387         hfs_find_exit(&fd);
388
389         return res;
390 }
391
392 int hfsplus_file_extend(struct inode *inode)
393 {
394         struct super_block *sb = inode->i_sb;
395         struct hfsplus_sb_info *sbi = HFSPLUS_SB(sb);
396         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
397         u32 start, len, goal;
398         int res;
399
400         if (sbi->alloc_file->i_size * 8 <
401             sbi->total_blocks - sbi->free_blocks + 8) {
402                 /* extend alloc file */
403                 printk(KERN_ERR "hfs: extend alloc file! "
404                                 "(%llu,%u,%u)\n",
405                         sbi->alloc_file->i_size * 8,
406                         sbi->total_blocks, sbi->free_blocks);
407                 return -ENOSPC;
408         }
409
410         mutex_lock(&hip->extents_lock);
411         if (hip->alloc_blocks == hip->first_blocks)
412                 goal = hfsplus_ext_lastblock(hip->first_extents);
413         else {
414                 res = hfsplus_ext_read_extent(inode, hip->alloc_blocks);
415                 if (res)
416                         goto out;
417                 goal = hfsplus_ext_lastblock(hip->cached_extents);
418         }
419
420         len = hip->clump_blocks;
421         start = hfsplus_block_allocate(sb, sbi->total_blocks, goal, &len);
422         if (start >= sbi->total_blocks) {
423                 start = hfsplus_block_allocate(sb, goal, 0, &len);
424                 if (start >= goal) {
425                         res = -ENOSPC;
426                         goto out;
427                 }
428         }
429
430         dprint(DBG_EXTENT, "extend %lu: %u,%u\n", inode->i_ino, start, len);
431
432         if (hip->alloc_blocks <= hip->first_blocks) {
433                 if (!hip->first_blocks) {
434                         dprint(DBG_EXTENT, "first extents\n");
435                         /* no extents yet */
436                         hip->first_extents[0].start_block = cpu_to_be32(start);
437                         hip->first_extents[0].block_count = cpu_to_be32(len);
438                         res = 0;
439                 } else {
440                         /* try to append to extents in inode */
441                         res = hfsplus_add_extent(hip->first_extents,
442                                                  hip->alloc_blocks,
443                                                  start, len);
444                         if (res == -ENOSPC)
445                                 goto insert_extent;
446                 }
447                 if (!res) {
448                         hfsplus_dump_extent(hip->first_extents);
449                         hip->first_blocks += len;
450                 }
451         } else {
452                 res = hfsplus_add_extent(hip->cached_extents,
453                                          hip->alloc_blocks - hip->cached_start,
454                                          start, len);
455                 if (!res) {
456                         hfsplus_dump_extent(hip->cached_extents);
457                         hip->extent_state |= HFSPLUS_EXT_DIRTY;
458                         hip->cached_blocks += len;
459                 } else if (res == -ENOSPC)
460                         goto insert_extent;
461         }
462 out:
463         mutex_unlock(&hip->extents_lock);
464         if (!res) {
465                 hip->alloc_blocks += len;
466                 hfsplus_mark_inode_dirty(inode, HFSPLUS_I_ALLOC_DIRTY);
467         }
468         return res;
469
470 insert_extent:
471         dprint(DBG_EXTENT, "insert new extent\n");
472         hfsplus_ext_write_extent_locked(inode);
473
474         memset(hip->cached_extents, 0, sizeof(hfsplus_extent_rec));
475         hip->cached_extents[0].start_block = cpu_to_be32(start);
476         hip->cached_extents[0].block_count = cpu_to_be32(len);
477         hfsplus_dump_extent(hip->cached_extents);
478         hip->extent_state |= HFSPLUS_EXT_DIRTY | HFSPLUS_EXT_NEW;
479         hip->cached_start = hip->alloc_blocks;
480         hip->cached_blocks = len;
481
482         res = 0;
483         goto out;
484 }
485
486 void hfsplus_file_truncate(struct inode *inode)
487 {
488         struct super_block *sb = inode->i_sb;
489         struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
490         struct hfs_find_data fd;
491         u32 alloc_cnt, blk_cnt, start;
492         int res;
493
494         dprint(DBG_INODE, "truncate: %lu, %llu -> %llu\n",
495                 inode->i_ino, (long long)hip->phys_size,
496                 inode->i_size);
497
498         if (inode->i_size > hip->phys_size) {
499                 struct address_space *mapping = inode->i_mapping;
500                 struct page *page;
501                 void *fsdata;
502                 u32 size = inode->i_size;
503                 int res;
504
505                 res = pagecache_write_begin(NULL, mapping, size, 0,
506                                                 AOP_FLAG_UNINTERRUPTIBLE,
507                                                 &page, &fsdata);
508                 if (res)
509                         return;
510                 res = pagecache_write_end(NULL, mapping, size,
511                         0, 0, page, fsdata);
512                 if (res < 0)
513                         return;
514                 mark_inode_dirty(inode);
515                 return;
516         } else if (inode->i_size == hip->phys_size)
517                 return;
518
519         blk_cnt = (inode->i_size + HFSPLUS_SB(sb)->alloc_blksz - 1) >>
520                         HFSPLUS_SB(sb)->alloc_blksz_shift;
521         alloc_cnt = hip->alloc_blocks;
522         if (blk_cnt == alloc_cnt)
523                 goto out;
524
525         mutex_lock(&hip->extents_lock);
526         hfs_find_init(HFSPLUS_SB(sb)->ext_tree, &fd);
527         while (1) {
528                 if (alloc_cnt == hip->first_blocks) {
529                         hfsplus_free_extents(sb, hip->first_extents,
530                                              alloc_cnt, alloc_cnt - blk_cnt);
531                         hfsplus_dump_extent(hip->first_extents);
532                         hip->first_blocks = blk_cnt;
533                         break;
534                 }
535                 res = __hfsplus_ext_cache_extent(&fd, inode, alloc_cnt);
536                 if (res)
537                         break;
538                 start = hip->cached_start;
539                 hfsplus_free_extents(sb, hip->cached_extents,
540                                      alloc_cnt - start, alloc_cnt - blk_cnt);
541                 hfsplus_dump_extent(hip->cached_extents);
542                 if (blk_cnt > start) {
543                         hip->extent_state |= HFSPLUS_EXT_DIRTY;
544                         break;
545                 }
546                 alloc_cnt = start;
547                 hip->cached_start = hip->cached_blocks = 0;
548                 hip->extent_state &= ~(HFSPLUS_EXT_DIRTY | HFSPLUS_EXT_NEW);
549                 hfs_brec_remove(&fd);
550         }
551         hfs_find_exit(&fd);
552         mutex_unlock(&hip->extents_lock);
553
554         hip->alloc_blocks = blk_cnt;
555 out:
556         hip->phys_size = inode->i_size;
557         hip->fs_blocks = (inode->i_size + sb->s_blocksize - 1) >>
558                 sb->s_blocksize_bits;
559         inode_set_bytes(inode, hip->fs_blocks << sb->s_blocksize_bits);
560         hfsplus_mark_inode_dirty(inode, HFSPLUS_I_ALLOC_DIRTY);
561 }