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