tail_conversion.c revision 0222e6571c332563a48d4cf5487b67feabe60b5e
1/*
2 * Copyright 1999 Hans Reiser, see reiserfs/README for licensing and copyright details
3 */
4
5#include <linux/time.h>
6#include <linux/pagemap.h>
7#include <linux/buffer_head.h>
8#include <linux/reiserfs_fs.h>
9
10/* access to tail : when one is going to read tail it must make sure, that is not running.
11 direct2indirect and indirect2direct can not run concurrently */
12
13/* Converts direct items to an unformatted node. Panics if file has no
14   tail. -ENOSPC if no disk space for conversion */
15/* path points to first direct item of the file regarless of how many of
16   them are there */
17int direct2indirect(struct reiserfs_transaction_handle *th, struct inode *inode,
18		    struct treepath *path, struct buffer_head *unbh,
19		    loff_t tail_offset)
20{
21	struct super_block *sb = inode->i_sb;
22	struct buffer_head *up_to_date_bh;
23	struct item_head *p_le_ih = PATH_PITEM_HEAD(path);
24	unsigned long total_tail = 0;
25	struct cpu_key end_key;	/* Key to search for the last byte of the
26				   converted item. */
27	struct item_head ind_ih;	/* new indirect item to be inserted or
28					   key of unfm pointer to be pasted */
29	int n_blk_size, n_retval;	/* returned value for reiserfs_insert_item and clones */
30	unp_t unfm_ptr;		/* Handle on an unformatted node
31				   that will be inserted in the
32				   tree. */
33
34	BUG_ON(!th->t_trans_id);
35
36	REISERFS_SB(sb)->s_direct2indirect++;
37
38	n_blk_size = sb->s_blocksize;
39
40	/* and key to search for append or insert pointer to the new
41	   unformatted node. */
42	copy_item_head(&ind_ih, p_le_ih);
43	set_le_ih_k_offset(&ind_ih, tail_offset);
44	set_le_ih_k_type(&ind_ih, TYPE_INDIRECT);
45
46	/* Set the key to search for the place for new unfm pointer */
47	make_cpu_key(&end_key, inode, tail_offset, TYPE_INDIRECT, 4);
48
49	/* FIXME: we could avoid this */
50	if (search_for_position_by_key(sb, &end_key, path) == POSITION_FOUND) {
51		reiserfs_error(sb, "PAP-14030",
52			       "pasted or inserted byte exists in "
53			       "the tree %K. Use fsck to repair.", &end_key);
54		pathrelse(path);
55		return -EIO;
56	}
57
58	p_le_ih = PATH_PITEM_HEAD(path);
59
60	unfm_ptr = cpu_to_le32(unbh->b_blocknr);
61
62	if (is_statdata_le_ih(p_le_ih)) {
63		/* Insert new indirect item. */
64		set_ih_free_space(&ind_ih, 0);	/* delete at nearest future */
65		put_ih_item_len(&ind_ih, UNFM_P_SIZE);
66		PATH_LAST_POSITION(path)++;
67		n_retval =
68		    reiserfs_insert_item(th, path, &end_key, &ind_ih, inode,
69					 (char *)&unfm_ptr);
70	} else {
71		/* Paste into last indirect item of an object. */
72		n_retval = reiserfs_paste_into_item(th, path, &end_key, inode,
73						    (char *)&unfm_ptr,
74						    UNFM_P_SIZE);
75	}
76	if (n_retval) {
77		return n_retval;
78	}
79	// note: from here there are two keys which have matching first
80	// three key components. They only differ by the fourth one.
81
82	/* Set the key to search for the direct items of the file */
83	make_cpu_key(&end_key, inode, max_reiserfs_offset(inode), TYPE_DIRECT,
84		     4);
85
86	/* Move bytes from the direct items to the new unformatted node
87	   and delete them. */
88	while (1) {
89		int tail_size;
90
91		/* end_key.k_offset is set so, that we will always have found
92		   last item of the file */
93		if (search_for_position_by_key(sb, &end_key, path) ==
94		    POSITION_FOUND)
95			reiserfs_panic(sb, "PAP-14050",
96				       "direct item (%K) not found", &end_key);
97		p_le_ih = PATH_PITEM_HEAD(path);
98		RFALSE(!is_direct_le_ih(p_le_ih),
99		       "vs-14055: direct item expected(%K), found %h",
100		       &end_key, p_le_ih);
101		tail_size = (le_ih_k_offset(p_le_ih) & (n_blk_size - 1))
102		    + ih_item_len(p_le_ih) - 1;
103
104		/* we only send the unbh pointer if the buffer is not up to date.
105		 ** this avoids overwriting good data from writepage() with old data
106		 ** from the disk or buffer cache
107		 ** Special case: unbh->b_page will be NULL if we are coming through
108		 ** DIRECT_IO handler here.
109		 */
110		if (!unbh->b_page || buffer_uptodate(unbh)
111		    || PageUptodate(unbh->b_page)) {
112			up_to_date_bh = NULL;
113		} else {
114			up_to_date_bh = unbh;
115		}
116		n_retval = reiserfs_delete_item(th, path, &end_key, inode,
117						up_to_date_bh);
118
119		total_tail += n_retval;
120		if (tail_size == n_retval)
121			// done: file does not have direct items anymore
122			break;
123
124	}
125	/* if we've copied bytes from disk into the page, we need to zero
126	 ** out the unused part of the block (it was not up to date before)
127	 */
128	if (up_to_date_bh) {
129		unsigned pgoff =
130		    (tail_offset + total_tail - 1) & (PAGE_CACHE_SIZE - 1);
131		char *kaddr = kmap_atomic(up_to_date_bh->b_page, KM_USER0);
132		memset(kaddr + pgoff, 0, n_blk_size - total_tail);
133		kunmap_atomic(kaddr, KM_USER0);
134	}
135
136	REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
137
138	return 0;
139}
140
141/* stolen from fs/buffer.c */
142void reiserfs_unmap_buffer(struct buffer_head *bh)
143{
144	lock_buffer(bh);
145	if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
146		BUG();
147	}
148	clear_buffer_dirty(bh);
149	/* Remove the buffer from whatever list it belongs to. We are mostly
150	   interested in removing it from per-sb j_dirty_buffers list, to avoid
151	   BUG() on attempt to write not mapped buffer */
152	if ((!list_empty(&bh->b_assoc_buffers) || bh->b_private) && bh->b_page) {
153		struct inode *inode = bh->b_page->mapping->host;
154		struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
155		spin_lock(&j->j_dirty_buffers_lock);
156		list_del_init(&bh->b_assoc_buffers);
157		reiserfs_free_jh(bh);
158		spin_unlock(&j->j_dirty_buffers_lock);
159	}
160	clear_buffer_mapped(bh);
161	clear_buffer_req(bh);
162	clear_buffer_new(bh);
163	bh->b_bdev = NULL;
164	unlock_buffer(bh);
165}
166
167/* this first locks inode (neither reads nor sync are permitted),
168   reads tail through page cache, insert direct item. When direct item
169   inserted successfully inode is left locked. Return value is always
170   what we expect from it (number of cut bytes). But when tail remains
171   in the unformatted node, we set mode to SKIP_BALANCING and unlock
172   inode */
173int indirect2direct(struct reiserfs_transaction_handle *th, struct inode *p_s_inode, struct page *page, struct treepath *p_s_path,	/* path to the indirect item. */
174		    const struct cpu_key *p_s_item_key,	/* Key to look for unformatted node pointer to be cut. */
175		    loff_t n_new_file_size,	/* New file size. */
176		    char *p_c_mode)
177{
178	struct super_block *p_s_sb = p_s_inode->i_sb;
179	struct item_head s_ih;
180	unsigned long n_block_size = p_s_sb->s_blocksize;
181	char *tail;
182	int tail_len, round_tail_len;
183	loff_t pos, pos1;	/* position of first byte of the tail */
184	struct cpu_key key;
185
186	BUG_ON(!th->t_trans_id);
187
188	REISERFS_SB(p_s_sb)->s_indirect2direct++;
189
190	*p_c_mode = M_SKIP_BALANCING;
191
192	/* store item head path points to. */
193	copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
194
195	tail_len = (n_new_file_size & (n_block_size - 1));
196	if (get_inode_sd_version(p_s_inode) == STAT_DATA_V2)
197		round_tail_len = ROUND_UP(tail_len);
198	else
199		round_tail_len = tail_len;
200
201	pos =
202	    le_ih_k_offset(&s_ih) - 1 + (ih_item_len(&s_ih) / UNFM_P_SIZE -
203					 1) * p_s_sb->s_blocksize;
204	pos1 = pos;
205
206	// we are protected by i_mutex. The tail can not disapper, not
207	// append can be done either
208	// we are in truncate or packing tail in file_release
209
210	tail = (char *)kmap(page);	/* this can schedule */
211
212	if (path_changed(&s_ih, p_s_path)) {
213		/* re-search indirect item */
214		if (search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path)
215		    == POSITION_NOT_FOUND)
216			reiserfs_panic(p_s_sb, "PAP-5520",
217				       "item to be converted %K does not exist",
218				       p_s_item_key);
219		copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
220#ifdef CONFIG_REISERFS_CHECK
221		pos = le_ih_k_offset(&s_ih) - 1 +
222		    (ih_item_len(&s_ih) / UNFM_P_SIZE -
223		     1) * p_s_sb->s_blocksize;
224		if (pos != pos1)
225			reiserfs_panic(p_s_sb, "vs-5530", "tail position "
226				       "changed while we were reading it");
227#endif
228	}
229
230	/* Set direct item header to insert. */
231	make_le_item_head(&s_ih, NULL, get_inode_item_key_version(p_s_inode),
232			  pos1 + 1, TYPE_DIRECT, round_tail_len,
233			  0xffff /*ih_free_space */ );
234
235	/* we want a pointer to the first byte of the tail in the page.
236	 ** the page was locked and this part of the page was up to date when
237	 ** indirect2direct was called, so we know the bytes are still valid
238	 */
239	tail = tail + (pos & (PAGE_CACHE_SIZE - 1));
240
241	PATH_LAST_POSITION(p_s_path)++;
242
243	key = *p_s_item_key;
244	set_cpu_key_k_type(&key, TYPE_DIRECT);
245	key.key_length = 4;
246	/* Insert tail as new direct item in the tree */
247	if (reiserfs_insert_item(th, p_s_path, &key, &s_ih, p_s_inode,
248				 tail ? tail : NULL) < 0) {
249		/* No disk memory. So we can not convert last unformatted node
250		   to the direct item.  In this case we used to adjust
251		   indirect items's ih_free_space. Now ih_free_space is not
252		   used, it would be ideal to write zeros to corresponding
253		   unformatted node. For now i_size is considered as guard for
254		   going out of file size */
255		kunmap(page);
256		return n_block_size - round_tail_len;
257	}
258	kunmap(page);
259
260	/* make sure to get the i_blocks changes from reiserfs_insert_item */
261	reiserfs_update_sd(th, p_s_inode);
262
263	// note: we have now the same as in above direct2indirect
264	// conversion: there are two keys which have matching first three
265	// key components. They only differ by the fouhth one.
266
267	/* We have inserted new direct item and must remove last
268	   unformatted node. */
269	*p_c_mode = M_CUT;
270
271	/* we store position of first direct item in the in-core inode */
272	//mark_file_with_tail (p_s_inode, pos1 + 1);
273	REISERFS_I(p_s_inode)->i_first_direct_byte = pos1 + 1;
274
275	return n_block_size - round_tail_len;
276}
277