Fix congestion_wait() sync/async vs read/write confusion
[linux-2.6] / fs / omfs / file.c
1 /*
2  * OMFS (as used by RIO Karma) file operations.
3  * Copyright (C) 2005 Bob Copeland <me@bobcopeland.com>
4  * Released under GPL v2.
5  */
6
7 #include <linux/version.h>
8 #include <linux/module.h>
9 #include <linux/fs.h>
10 #include <linux/buffer_head.h>
11 #include <linux/mpage.h>
12 #include "omfs.h"
13
14 static u32 omfs_max_extents(struct omfs_sb_info *sbi, int offset)
15 {
16         return (sbi->s_sys_blocksize - offset -
17                 sizeof(struct omfs_extent)) /
18                 sizeof(struct omfs_extent_entry) + 1;
19 }
20
21 void omfs_make_empty_table(struct buffer_head *bh, int offset)
22 {
23         struct omfs_extent *oe = (struct omfs_extent *) &bh->b_data[offset];
24
25         oe->e_next = ~cpu_to_be64(0ULL);
26         oe->e_extent_count = cpu_to_be32(1),
27         oe->e_fill = cpu_to_be32(0x22),
28         oe->e_entry.e_cluster = ~cpu_to_be64(0ULL);
29         oe->e_entry.e_blocks = ~cpu_to_be64(0ULL);
30 }
31
32 int omfs_shrink_inode(struct inode *inode)
33 {
34         struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb);
35         struct omfs_extent *oe;
36         struct omfs_extent_entry *entry;
37         struct buffer_head *bh;
38         u64 next, last;
39         u32 extent_count;
40         u32 max_extents;
41         int ret;
42
43         /* traverse extent table, freeing each entry that is greater
44          * than inode->i_size;
45          */
46         next = inode->i_ino;
47
48         /* only support truncate -> 0 for now */
49         ret = -EIO;
50         if (inode->i_size != 0)
51                 goto out;
52
53         bh = sb_bread(inode->i_sb, clus_to_blk(sbi, next));
54         if (!bh)
55                 goto out;
56
57         oe = (struct omfs_extent *)(&bh->b_data[OMFS_EXTENT_START]);
58         max_extents = omfs_max_extents(sbi, OMFS_EXTENT_START);
59
60         for (;;) {
61
62                 if (omfs_is_bad(sbi, (struct omfs_header *) bh->b_data, next))
63                         goto out_brelse;
64
65                 extent_count = be32_to_cpu(oe->e_extent_count);
66
67                 if (extent_count > max_extents)
68                         goto out_brelse;
69
70                 last = next;
71                 next = be64_to_cpu(oe->e_next);
72                 entry = &oe->e_entry;
73
74                 /* ignore last entry as it is the terminator */
75                 for (; extent_count > 1; extent_count--) {
76                         u64 start, count;
77                         start = be64_to_cpu(entry->e_cluster);
78                         count = be64_to_cpu(entry->e_blocks);
79
80                         omfs_clear_range(inode->i_sb, start, (int) count);
81                         entry++;
82                 }
83                 omfs_make_empty_table(bh, (char *) oe - bh->b_data);
84                 mark_buffer_dirty(bh);
85                 brelse(bh);
86
87                 if (last != inode->i_ino)
88                         omfs_clear_range(inode->i_sb, last, sbi->s_mirrors);
89
90                 if (next == ~0)
91                         break;
92
93                 bh = sb_bread(inode->i_sb, clus_to_blk(sbi, next));
94                 if (!bh)
95                         goto out;
96                 oe = (struct omfs_extent *) (&bh->b_data[OMFS_EXTENT_CONT]);
97                 max_extents = omfs_max_extents(sbi, OMFS_EXTENT_CONT);
98         }
99         ret = 0;
100 out:
101         return ret;
102 out_brelse:
103         brelse(bh);
104         return ret;
105 }
106
107 static void omfs_truncate(struct inode *inode)
108 {
109         omfs_shrink_inode(inode);
110         mark_inode_dirty(inode);
111 }
112
113 /*
114  * Add new blocks to the current extent, or create new entries/continuations
115  * as necessary.
116  */
117 static int omfs_grow_extent(struct inode *inode, struct omfs_extent *oe,
118                         u64 *ret_block)
119 {
120         struct omfs_extent_entry *terminator;
121         struct omfs_extent_entry *entry = &oe->e_entry;
122         struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb);
123         u32 extent_count = be32_to_cpu(oe->e_extent_count);
124         u64 new_block = 0;
125         u32 max_count;
126         int new_count;
127         int ret = 0;
128
129         /* reached the end of the extent table with no blocks mapped.
130          * there are three possibilities for adding: grow last extent,
131          * add a new extent to the current extent table, and add a
132          * continuation inode.  in last two cases need an allocator for
133          * sbi->s_cluster_size
134          */
135
136         /* TODO: handle holes */
137
138         /* should always have a terminator */
139         if (extent_count < 1)
140                 return -EIO;
141
142         /* trivially grow current extent, if next block is not taken */
143         terminator = entry + extent_count - 1;
144         if (extent_count > 1) {
145                 entry = terminator-1;
146                 new_block = be64_to_cpu(entry->e_cluster) +
147                         be64_to_cpu(entry->e_blocks);
148
149                 if (omfs_allocate_block(inode->i_sb, new_block)) {
150                         entry->e_blocks =
151                                 cpu_to_be64(be64_to_cpu(entry->e_blocks) + 1);
152                         terminator->e_blocks = ~(cpu_to_be64(
153                                 be64_to_cpu(~terminator->e_blocks) + 1));
154                         goto out;
155                 }
156         }
157         max_count = omfs_max_extents(sbi, OMFS_EXTENT_START);
158
159         /* TODO: add a continuation block here */
160         if (be32_to_cpu(oe->e_extent_count) > max_count-1)
161                 return -EIO;
162
163         /* try to allocate a new cluster */
164         ret = omfs_allocate_range(inode->i_sb, 1, sbi->s_clustersize,
165                 &new_block, &new_count);
166         if (ret)
167                 goto out_fail;
168
169         /* copy terminator down an entry */
170         entry = terminator;
171         terminator++;
172         memcpy(terminator, entry, sizeof(struct omfs_extent_entry));
173
174         entry->e_cluster = cpu_to_be64(new_block);
175         entry->e_blocks = cpu_to_be64((u64) new_count);
176
177         terminator->e_blocks = ~(cpu_to_be64(
178                 be64_to_cpu(~terminator->e_blocks) + (u64) new_count));
179
180         /* write in new entry */
181         oe->e_extent_count = cpu_to_be32(1 + be32_to_cpu(oe->e_extent_count));
182
183 out:
184         *ret_block = new_block;
185 out_fail:
186         return ret;
187 }
188
189 /*
190  * Scans across the directory table for a given file block number.
191  * If block not found, return 0.
192  */
193 static sector_t find_block(struct inode *inode, struct omfs_extent_entry *ent,
194                         sector_t block, int count, int *left)
195 {
196         /* count > 1 because of terminator */
197         sector_t searched = 0;
198         for (; count > 1; count--) {
199                 int numblocks = clus_to_blk(OMFS_SB(inode->i_sb),
200                         be64_to_cpu(ent->e_blocks));
201
202                 if (block >= searched  &&
203                     block < searched + numblocks) {
204                         /*
205                          * found it at cluster + (block - searched)
206                          * numblocks - (block - searched) is remainder
207                          */
208                         *left = numblocks - (block - searched);
209                         return clus_to_blk(OMFS_SB(inode->i_sb),
210                                 be64_to_cpu(ent->e_cluster)) +
211                                 block - searched;
212                 }
213                 searched += numblocks;
214                 ent++;
215         }
216         return 0;
217 }
218
219 static int omfs_get_block(struct inode *inode, sector_t block,
220                           struct buffer_head *bh_result, int create)
221 {
222         struct buffer_head *bh;
223         sector_t next, offset;
224         int ret;
225         u64 new_block;
226         u32 max_extents;
227         int extent_count;
228         struct omfs_extent *oe;
229         struct omfs_extent_entry *entry;
230         struct omfs_sb_info *sbi = OMFS_SB(inode->i_sb);
231         int max_blocks = bh_result->b_size >> inode->i_blkbits;
232         int remain;
233
234         ret = -EIO;
235         bh = sb_bread(inode->i_sb, clus_to_blk(sbi, inode->i_ino));
236         if (!bh)
237                 goto out;
238
239         oe = (struct omfs_extent *)(&bh->b_data[OMFS_EXTENT_START]);
240         max_extents = omfs_max_extents(sbi, OMFS_EXTENT_START);
241         next = inode->i_ino;
242
243         for (;;) {
244
245                 if (omfs_is_bad(sbi, (struct omfs_header *) bh->b_data, next))
246                         goto out_brelse;
247
248                 extent_count = be32_to_cpu(oe->e_extent_count);
249                 next = be64_to_cpu(oe->e_next);
250                 entry = &oe->e_entry;
251
252                 if (extent_count > max_extents)
253                         goto out_brelse;
254
255                 offset = find_block(inode, entry, block, extent_count, &remain);
256                 if (offset > 0) {
257                         ret = 0;
258                         map_bh(bh_result, inode->i_sb, offset);
259                         if (remain > max_blocks)
260                                 remain = max_blocks;
261                         bh_result->b_size = (remain << inode->i_blkbits);
262                         goto out_brelse;
263                 }
264                 if (next == ~0)
265                         break;
266
267                 brelse(bh);
268                 bh = sb_bread(inode->i_sb, clus_to_blk(sbi, next));
269                 if (!bh)
270                         goto out;
271                 oe = (struct omfs_extent *) (&bh->b_data[OMFS_EXTENT_CONT]);
272                 max_extents = omfs_max_extents(sbi, OMFS_EXTENT_CONT);
273         }
274         if (create) {
275                 ret = omfs_grow_extent(inode, oe, &new_block);
276                 if (ret == 0) {
277                         mark_buffer_dirty(bh);
278                         mark_inode_dirty(inode);
279                         map_bh(bh_result, inode->i_sb,
280                                         clus_to_blk(sbi, new_block));
281                 }
282         }
283 out_brelse:
284         brelse(bh);
285 out:
286         return ret;
287 }
288
289 static int omfs_readpage(struct file *file, struct page *page)
290 {
291         return block_read_full_page(page, omfs_get_block);
292 }
293
294 static int omfs_readpages(struct file *file, struct address_space *mapping,
295                 struct list_head *pages, unsigned nr_pages)
296 {
297         return mpage_readpages(mapping, pages, nr_pages, omfs_get_block);
298 }
299
300 static int omfs_writepage(struct page *page, struct writeback_control *wbc)
301 {
302         return block_write_full_page(page, omfs_get_block, wbc);
303 }
304
305 static int
306 omfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
307 {
308         return mpage_writepages(mapping, wbc, omfs_get_block);
309 }
310
311 static int omfs_write_begin(struct file *file, struct address_space *mapping,
312                         loff_t pos, unsigned len, unsigned flags,
313                         struct page **pagep, void **fsdata)
314 {
315         *pagep = NULL;
316         return block_write_begin(file, mapping, pos, len, flags,
317                                 pagep, fsdata, omfs_get_block);
318 }
319
320 static sector_t omfs_bmap(struct address_space *mapping, sector_t block)
321 {
322         return generic_block_bmap(mapping, block, omfs_get_block);
323 }
324
325 struct file_operations omfs_file_operations = {
326         .llseek = generic_file_llseek,
327         .read = do_sync_read,
328         .write = do_sync_write,
329         .aio_read = generic_file_aio_read,
330         .aio_write = generic_file_aio_write,
331         .mmap = generic_file_mmap,
332         .fsync = simple_fsync,
333         .splice_read = generic_file_splice_read,
334 };
335
336 struct inode_operations omfs_file_inops = {
337         .truncate = omfs_truncate
338 };
339
340 struct address_space_operations omfs_aops = {
341         .readpage = omfs_readpage,
342         .readpages = omfs_readpages,
343         .writepage = omfs_writepage,
344         .writepages = omfs_writepages,
345         .sync_page = block_sync_page,
346         .write_begin = omfs_write_begin,
347         .write_end = generic_write_end,
348         .bmap = omfs_bmap,
349 };
350