Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/j.anaszewski...
[linux-2.6-block.git] / fs / reiserfs / tail_conversion.c
... / ...
CommitLineData
1/*
2 * Copyright 1999 Hans Reiser, see reiserfs/README for licensing and copyright
3 * details
4 */
5
6#include <linux/time.h>
7#include <linux/pagemap.h>
8#include <linux/buffer_head.h>
9#include "reiserfs.h"
10
11/*
12 * access to tail : when one is going to read tail it must make sure, that is
13 * not running. direct2indirect and indirect2direct can not run concurrently
14 */
15
16/*
17 * Converts direct items to an unformatted node. Panics if file has no
18 * tail. -ENOSPC if no disk space for conversion
19 */
20/*
21 * path points to first direct item of the file regardless of how many of
22 * them are there
23 */
24int direct2indirect(struct reiserfs_transaction_handle *th, struct inode *inode,
25 struct treepath *path, struct buffer_head *unbh,
26 loff_t tail_offset)
27{
28 struct super_block *sb = inode->i_sb;
29 struct buffer_head *up_to_date_bh;
30 struct item_head *p_le_ih = tp_item_head(path);
31 unsigned long total_tail = 0;
32
33 /* Key to search for the last byte of the converted item. */
34 struct cpu_key end_key;
35
36 /*
37 * new indirect item to be inserted or key
38 * of unfm pointer to be pasted
39 */
40 struct item_head ind_ih;
41 int blk_size;
42 /* returned value for reiserfs_insert_item and clones */
43 int retval;
44 /* Handle on an unformatted node that will be inserted in the tree. */
45 unp_t unfm_ptr;
46
47 BUG_ON(!th->t_trans_id);
48
49 REISERFS_SB(sb)->s_direct2indirect++;
50
51 blk_size = sb->s_blocksize;
52
53 /*
54 * and key to search for append or insert pointer to the new
55 * unformatted node.
56 */
57 copy_item_head(&ind_ih, p_le_ih);
58 set_le_ih_k_offset(&ind_ih, tail_offset);
59 set_le_ih_k_type(&ind_ih, TYPE_INDIRECT);
60
61 /* Set the key to search for the place for new unfm pointer */
62 make_cpu_key(&end_key, inode, tail_offset, TYPE_INDIRECT, 4);
63
64 /* FIXME: we could avoid this */
65 if (search_for_position_by_key(sb, &end_key, path) == POSITION_FOUND) {
66 reiserfs_error(sb, "PAP-14030",
67 "pasted or inserted byte exists in "
68 "the tree %K. Use fsck to repair.", &end_key);
69 pathrelse(path);
70 return -EIO;
71 }
72
73 p_le_ih = tp_item_head(path);
74
75 unfm_ptr = cpu_to_le32(unbh->b_blocknr);
76
77 if (is_statdata_le_ih(p_le_ih)) {
78 /* Insert new indirect item. */
79 set_ih_free_space(&ind_ih, 0); /* delete at nearest future */
80 put_ih_item_len(&ind_ih, UNFM_P_SIZE);
81 PATH_LAST_POSITION(path)++;
82 retval =
83 reiserfs_insert_item(th, path, &end_key, &ind_ih, inode,
84 (char *)&unfm_ptr);
85 } else {
86 /* Paste into last indirect item of an object. */
87 retval = reiserfs_paste_into_item(th, path, &end_key, inode,
88 (char *)&unfm_ptr,
89 UNFM_P_SIZE);
90 }
91 if (retval) {
92 return retval;
93 }
94 /*
95 * note: from here there are two keys which have matching first
96 * three key components. They only differ by the fourth one.
97 */
98
99 /* Set the key to search for the direct items of the file */
100 make_cpu_key(&end_key, inode, max_reiserfs_offset(inode), TYPE_DIRECT,
101 4);
102
103 /*
104 * Move bytes from the direct items to the new unformatted node
105 * and delete them.
106 */
107 while (1) {
108 int tail_size;
109
110 /*
111 * end_key.k_offset is set so, that we will always have found
112 * last item of the file
113 */
114 if (search_for_position_by_key(sb, &end_key, path) ==
115 POSITION_FOUND)
116 reiserfs_panic(sb, "PAP-14050",
117 "direct item (%K) not found", &end_key);
118 p_le_ih = tp_item_head(path);
119 RFALSE(!is_direct_le_ih(p_le_ih),
120 "vs-14055: direct item expected(%K), found %h",
121 &end_key, p_le_ih);
122 tail_size = (le_ih_k_offset(p_le_ih) & (blk_size - 1))
123 + ih_item_len(p_le_ih) - 1;
124
125 /*
126 * we only send the unbh pointer if the buffer is not
127 * up to date. this avoids overwriting good data from
128 * writepage() with old data from the disk or buffer cache
129 * Special case: unbh->b_page will be NULL if we are coming
130 * through DIRECT_IO handler here.
131 */
132 if (!unbh->b_page || buffer_uptodate(unbh)
133 || PageUptodate(unbh->b_page)) {
134 up_to_date_bh = NULL;
135 } else {
136 up_to_date_bh = unbh;
137 }
138 retval = reiserfs_delete_item(th, path, &end_key, inode,
139 up_to_date_bh);
140
141 total_tail += retval;
142
143 /* done: file does not have direct items anymore */
144 if (tail_size == retval)
145 break;
146
147 }
148 /*
149 * if we've copied bytes from disk into the page, we need to zero
150 * out the unused part of the block (it was not up to date before)
151 */
152 if (up_to_date_bh) {
153 unsigned pgoff =
154 (tail_offset + total_tail - 1) & (PAGE_SIZE - 1);
155 char *kaddr = kmap_atomic(up_to_date_bh->b_page);
156 memset(kaddr + pgoff, 0, blk_size - total_tail);
157 kunmap_atomic(kaddr);
158 }
159
160 REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
161
162 return 0;
163}
164
165/* stolen from fs/buffer.c */
166void reiserfs_unmap_buffer(struct buffer_head *bh)
167{
168 lock_buffer(bh);
169 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
170 BUG();
171 }
172 clear_buffer_dirty(bh);
173 /*
174 * Remove the buffer from whatever list it belongs to. We are mostly
175 * interested in removing it from per-sb j_dirty_buffers list, to avoid
176 * BUG() on attempt to write not mapped buffer
177 */
178 if ((!list_empty(&bh->b_assoc_buffers) || bh->b_private) && bh->b_page) {
179 struct inode *inode = bh->b_page->mapping->host;
180 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
181 spin_lock(&j->j_dirty_buffers_lock);
182 list_del_init(&bh->b_assoc_buffers);
183 reiserfs_free_jh(bh);
184 spin_unlock(&j->j_dirty_buffers_lock);
185 }
186 clear_buffer_mapped(bh);
187 clear_buffer_req(bh);
188 clear_buffer_new(bh);
189 bh->b_bdev = NULL;
190 unlock_buffer(bh);
191}
192
193/*
194 * this first locks inode (neither reads nor sync are permitted),
195 * reads tail through page cache, insert direct item. When direct item
196 * inserted successfully inode is left locked. Return value is always
197 * what we expect from it (number of cut bytes). But when tail remains
198 * in the unformatted node, we set mode to SKIP_BALANCING and unlock
199 * inode
200 */
201int indirect2direct(struct reiserfs_transaction_handle *th,
202 struct inode *inode, struct page *page,
203 struct treepath *path, /* path to the indirect item. */
204 const struct cpu_key *item_key, /* Key to look for
205 * unformatted node
206 * pointer to be cut. */
207 loff_t n_new_file_size, /* New file size. */
208 char *mode)
209{
210 struct super_block *sb = inode->i_sb;
211 struct item_head s_ih;
212 unsigned long block_size = sb->s_blocksize;
213 char *tail;
214 int tail_len, round_tail_len;
215 loff_t pos, pos1; /* position of first byte of the tail */
216 struct cpu_key key;
217
218 BUG_ON(!th->t_trans_id);
219
220 REISERFS_SB(sb)->s_indirect2direct++;
221
222 *mode = M_SKIP_BALANCING;
223
224 /* store item head path points to. */
225 copy_item_head(&s_ih, tp_item_head(path));
226
227 tail_len = (n_new_file_size & (block_size - 1));
228 if (get_inode_sd_version(inode) == STAT_DATA_V2)
229 round_tail_len = ROUND_UP(tail_len);
230 else
231 round_tail_len = tail_len;
232
233 pos =
234 le_ih_k_offset(&s_ih) - 1 + (ih_item_len(&s_ih) / UNFM_P_SIZE -
235 1) * sb->s_blocksize;
236 pos1 = pos;
237
238 /*
239 * we are protected by i_mutex. The tail can not disapper, not
240 * append can be done either
241 * we are in truncate or packing tail in file_release
242 */
243
244 tail = (char *)kmap(page); /* this can schedule */
245
246 if (path_changed(&s_ih, path)) {
247 /* re-search indirect item */
248 if (search_for_position_by_key(sb, item_key, path)
249 == POSITION_NOT_FOUND)
250 reiserfs_panic(sb, "PAP-5520",
251 "item to be converted %K does not exist",
252 item_key);
253 copy_item_head(&s_ih, tp_item_head(path));
254#ifdef CONFIG_REISERFS_CHECK
255 pos = le_ih_k_offset(&s_ih) - 1 +
256 (ih_item_len(&s_ih) / UNFM_P_SIZE -
257 1) * sb->s_blocksize;
258 if (pos != pos1)
259 reiserfs_panic(sb, "vs-5530", "tail position "
260 "changed while we were reading it");
261#endif
262 }
263
264 /* Set direct item header to insert. */
265 make_le_item_head(&s_ih, NULL, get_inode_item_key_version(inode),
266 pos1 + 1, TYPE_DIRECT, round_tail_len,
267 0xffff /*ih_free_space */ );
268
269 /*
270 * we want a pointer to the first byte of the tail in the page.
271 * the page was locked and this part of the page was up to date when
272 * indirect2direct was called, so we know the bytes are still valid
273 */
274 tail = tail + (pos & (PAGE_SIZE - 1));
275
276 PATH_LAST_POSITION(path)++;
277
278 key = *item_key;
279 set_cpu_key_k_type(&key, TYPE_DIRECT);
280 key.key_length = 4;
281 /* Insert tail as new direct item in the tree */
282 if (reiserfs_insert_item(th, path, &key, &s_ih, inode,
283 tail ? tail : NULL) < 0) {
284 /*
285 * No disk memory. So we can not convert last unformatted node
286 * to the direct item. In this case we used to adjust
287 * indirect items's ih_free_space. Now ih_free_space is not
288 * used, it would be ideal to write zeros to corresponding
289 * unformatted node. For now i_size is considered as guard for
290 * going out of file size
291 */
292 kunmap(page);
293 return block_size - round_tail_len;
294 }
295 kunmap(page);
296
297 /* make sure to get the i_blocks changes from reiserfs_insert_item */
298 reiserfs_update_sd(th, inode);
299
300 /*
301 * note: we have now the same as in above direct2indirect
302 * conversion: there are two keys which have matching first three
303 * key components. They only differ by the fourth one.
304 */
305
306 /*
307 * We have inserted new direct item and must remove last
308 * unformatted node.
309 */
310 *mode = M_CUT;
311
312 /* we store position of first direct item in the in-core inode */
313 /* mark_file_with_tail (inode, pos1 + 1); */
314 REISERFS_I(inode)->i_first_direct_byte = pos1 + 1;
315
316 return block_size - round_tail_len;
317}