tail_conversion.c 9.1 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
/*
 * Copyright 1999 Hans Reiser, see reiserfs/README for licensing and copyright details
 */

#include <linux/time.h>
#include <linux/pagemap.h>
#include <linux/buffer_head.h>
#include <linux/reiserfs_fs.h>

/* access to tail : when one is going to read tail it must make sure, that is not running.
 direct2indirect and indirect2direct can not run concurrently */

/* Converts direct items to an unformatted node. Panics if file has no
   tail. -ENOSPC if no disk space for conversion */
/* path points to first direct item of the file regarless of how many of
   them are there */
17
int direct2indirect(struct reiserfs_transaction_handle *th, struct inode *inode,
18
		    struct treepath *path, struct buffer_head *unbh,
19
		    loff_t tail_offset)
L
Linus Torvalds 已提交
20
{
21 22 23 24 25 26 27 28
	struct super_block *sb = inode->i_sb;
	struct buffer_head *up_to_date_bh;
	struct item_head *p_le_ih = PATH_PITEM_HEAD(path);
	unsigned long total_tail = 0;
	struct cpu_key end_key;	/* Key to search for the last byte of the
				   converted item. */
	struct item_head ind_ih;	/* new indirect item to be inserted or
					   key of unfm pointer to be pasted */
J
Jeff Mahoney 已提交
29
	int blk_size, retval;	/* returned value for reiserfs_insert_item and clones */
30 31 32 33 34 35 36 37
	unp_t unfm_ptr;		/* Handle on an unformatted node
				   that will be inserted in the
				   tree. */

	BUG_ON(!th->t_trans_id);

	REISERFS_SB(sb)->s_direct2indirect++;

J
Jeff Mahoney 已提交
38
	blk_size = sb->s_blocksize;
39 40 41 42 43 44 45 46 47 48

	/* and key to search for append or insert pointer to the new
	   unformatted node. */
	copy_item_head(&ind_ih, p_le_ih);
	set_le_ih_k_offset(&ind_ih, tail_offset);
	set_le_ih_k_type(&ind_ih, TYPE_INDIRECT);

	/* Set the key to search for the place for new unfm pointer */
	make_cpu_key(&end_key, inode, tail_offset, TYPE_INDIRECT, 4);

49
	/* FIXME: we could avoid this */
50
	if (search_for_position_by_key(sb, &end_key, path) == POSITION_FOUND) {
J
Jeff Mahoney 已提交
51 52 53
		reiserfs_error(sb, "PAP-14030",
			       "pasted or inserted byte exists in "
			       "the tree %K. Use fsck to repair.", &end_key);
54 55 56 57 58 59 60 61 62 63 64 65 66
		pathrelse(path);
		return -EIO;
	}

	p_le_ih = PATH_PITEM_HEAD(path);

	unfm_ptr = cpu_to_le32(unbh->b_blocknr);

	if (is_statdata_le_ih(p_le_ih)) {
		/* Insert new indirect item. */
		set_ih_free_space(&ind_ih, 0);	/* delete at nearest future */
		put_ih_item_len(&ind_ih, UNFM_P_SIZE);
		PATH_LAST_POSITION(path)++;
J
Jeff Mahoney 已提交
67
		retval =
68
		    reiserfs_insert_item(th, path, &end_key, &ind_ih, inode,
L
Linus Torvalds 已提交
69 70
					 (char *)&unfm_ptr);
	} else {
71
		/* Paste into last indirect item of an object. */
J
Jeff Mahoney 已提交
72
		retval = reiserfs_paste_into_item(th, path, &end_key, inode,
73 74
						    (char *)&unfm_ptr,
						    UNFM_P_SIZE);
L
Linus Torvalds 已提交
75
	}
J
Jeff Mahoney 已提交
76 77
	if (retval) {
		return retval;
78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
	}
	// note: from here there are two keys which have matching first
	// three key components. They only differ by the fourth one.

	/* Set the key to search for the direct items of the file */
	make_cpu_key(&end_key, inode, max_reiserfs_offset(inode), TYPE_DIRECT,
		     4);

	/* Move bytes from the direct items to the new unformatted node
	   and delete them. */
	while (1) {
		int tail_size;

		/* end_key.k_offset is set so, that we will always have found
		   last item of the file */
		if (search_for_position_by_key(sb, &end_key, path) ==
		    POSITION_FOUND)
J
Jeff Mahoney 已提交
95
			reiserfs_panic(sb, "PAP-14050",
96 97 98 99 100
				       "direct item (%K) not found", &end_key);
		p_le_ih = PATH_PITEM_HEAD(path);
		RFALSE(!is_direct_le_ih(p_le_ih),
		       "vs-14055: direct item expected(%K), found %h",
		       &end_key, p_le_ih);
J
Jeff Mahoney 已提交
101
		tail_size = (le_ih_k_offset(p_le_ih) & (blk_size - 1))
102 103 104 105 106 107 108 109 110 111 112 113 114 115
		    + ih_item_len(p_le_ih) - 1;

		/* we only send the unbh pointer if the buffer is not up to date.
		 ** this avoids overwriting good data from writepage() with old data
		 ** from the disk or buffer cache
		 ** Special case: unbh->b_page will be NULL if we are coming through
		 ** DIRECT_IO handler here.
		 */
		if (!unbh->b_page || buffer_uptodate(unbh)
		    || PageUptodate(unbh->b_page)) {
			up_to_date_bh = NULL;
		} else {
			up_to_date_bh = unbh;
		}
J
Jeff Mahoney 已提交
116
		retval = reiserfs_delete_item(th, path, &end_key, inode,
117 118
						up_to_date_bh);

J
Jeff Mahoney 已提交
119 120
		total_tail += retval;
		if (tail_size == retval)
121 122
			// done: file does not have direct items anymore
			break;
L
Linus Torvalds 已提交
123

124 125 126 127 128 129 130 131
	}
	/* if we've copied bytes from disk into the page, we need to zero
	 ** out the unused part of the block (it was not up to date before)
	 */
	if (up_to_date_bh) {
		unsigned pgoff =
		    (tail_offset + total_tail - 1) & (PAGE_CACHE_SIZE - 1);
		char *kaddr = kmap_atomic(up_to_date_bh->b_page, KM_USER0);
J
Jeff Mahoney 已提交
132
		memset(kaddr + pgoff, 0, blk_size - total_tail);
133 134 135 136 137 138 139
		kunmap_atomic(kaddr, KM_USER0);
	}

	REISERFS_I(inode)->i_first_direct_byte = U32_MAX;

	return 0;
}
L
Linus Torvalds 已提交
140 141

/* stolen from fs/buffer.c */
142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
void reiserfs_unmap_buffer(struct buffer_head *bh)
{
	lock_buffer(bh);
	if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
		BUG();
	}
	clear_buffer_dirty(bh);
	/* Remove the buffer from whatever list it belongs to. We are mostly
	   interested in removing it from per-sb j_dirty_buffers list, to avoid
	   BUG() on attempt to write not mapped buffer */
	if ((!list_empty(&bh->b_assoc_buffers) || bh->b_private) && bh->b_page) {
		struct inode *inode = bh->b_page->mapping->host;
		struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
		spin_lock(&j->j_dirty_buffers_lock);
		list_del_init(&bh->b_assoc_buffers);
		reiserfs_free_jh(bh);
		spin_unlock(&j->j_dirty_buffers_lock);
	}
	clear_buffer_mapped(bh);
	clear_buffer_req(bh);
	clear_buffer_new(bh);
	bh->b_bdev = NULL;
	unlock_buffer(bh);
L
Linus Torvalds 已提交
165 166 167 168 169 170 171 172
}

/* this first locks inode (neither reads nor sync are permitted),
   reads tail through page cache, insert direct item. When direct item
   inserted successfully inode is left locked. Return value is always
   what we expect from it (number of cut bytes). But when tail remains
   in the unformatted node, we set mode to SKIP_BALANCING and unlock
   inode */
173 174
int indirect2direct(struct reiserfs_transaction_handle *th,
		    struct inode *inode, struct page *page,
J
Jeff Mahoney 已提交
175 176 177 178
		    struct treepath *path,	/* path to the indirect item. */
		    const struct cpu_key *item_key,	/* Key to look for
							 * unformatted node
							 * pointer to be cut. */
179
		    loff_t n_new_file_size,	/* New file size. */
J
Jeff Mahoney 已提交
180
		    char *mode)
L
Linus Torvalds 已提交
181
{
182
	struct super_block *sb = inode->i_sb;
183
	struct item_head s_ih;
J
Jeff Mahoney 已提交
184
	unsigned long block_size = sb->s_blocksize;
185 186 187 188
	char *tail;
	int tail_len, round_tail_len;
	loff_t pos, pos1;	/* position of first byte of the tail */
	struct cpu_key key;
L
Linus Torvalds 已提交
189

190
	BUG_ON(!th->t_trans_id);
L
Linus Torvalds 已提交
191

J
Jeff Mahoney 已提交
192
	REISERFS_SB(sb)->s_indirect2direct++;
L
Linus Torvalds 已提交
193

J
Jeff Mahoney 已提交
194
	*mode = M_SKIP_BALANCING;
L
Linus Torvalds 已提交
195

196
	/* store item head path points to. */
J
Jeff Mahoney 已提交
197
	copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
198

J
Jeff Mahoney 已提交
199
	tail_len = (n_new_file_size & (block_size - 1));
200
	if (get_inode_sd_version(inode) == STAT_DATA_V2)
201 202 203 204 205 206
		round_tail_len = ROUND_UP(tail_len);
	else
		round_tail_len = tail_len;

	pos =
	    le_ih_k_offset(&s_ih) - 1 + (ih_item_len(&s_ih) / UNFM_P_SIZE -
J
Jeff Mahoney 已提交
207
					 1) * sb->s_blocksize;
208 209
	pos1 = pos;

210
	// we are protected by i_mutex. The tail can not disapper, not
211 212 213 214 215
	// append can be done either
	// we are in truncate or packing tail in file_release

	tail = (char *)kmap(page);	/* this can schedule */

J
Jeff Mahoney 已提交
216
	if (path_changed(&s_ih, path)) {
217
		/* re-search indirect item */
J
Jeff Mahoney 已提交
218
		if (search_for_position_by_key(sb, item_key, path)
219
		    == POSITION_NOT_FOUND)
J
Jeff Mahoney 已提交
220
			reiserfs_panic(sb, "PAP-5520",
221
				       "item to be converted %K does not exist",
J
Jeff Mahoney 已提交
222 223
				       item_key);
		copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
L
Linus Torvalds 已提交
224
#ifdef CONFIG_REISERFS_CHECK
225 226
		pos = le_ih_k_offset(&s_ih) - 1 +
		    (ih_item_len(&s_ih) / UNFM_P_SIZE -
J
Jeff Mahoney 已提交
227
		     1) * sb->s_blocksize;
228
		if (pos != pos1)
J
Jeff Mahoney 已提交
229
			reiserfs_panic(sb, "vs-5530", "tail position "
J
Jeff Mahoney 已提交
230
				       "changed while we were reading it");
L
Linus Torvalds 已提交
231
#endif
232
	}
L
Linus Torvalds 已提交
233

234
	/* Set direct item header to insert. */
235
	make_le_item_head(&s_ih, NULL, get_inode_item_key_version(inode),
236 237 238 239 240 241 242 243 244
			  pos1 + 1, TYPE_DIRECT, round_tail_len,
			  0xffff /*ih_free_space */ );

	/* we want a pointer to the first byte of the tail in the page.
	 ** the page was locked and this part of the page was up to date when
	 ** indirect2direct was called, so we know the bytes are still valid
	 */
	tail = tail + (pos & (PAGE_CACHE_SIZE - 1));

J
Jeff Mahoney 已提交
245
	PATH_LAST_POSITION(path)++;
246

J
Jeff Mahoney 已提交
247
	key = *item_key;
248 249 250
	set_cpu_key_k_type(&key, TYPE_DIRECT);
	key.key_length = 4;
	/* Insert tail as new direct item in the tree */
J
Jeff Mahoney 已提交
251
	if (reiserfs_insert_item(th, path, &key, &s_ih, inode,
252 253 254 255 256 257 258 259
				 tail ? tail : NULL) < 0) {
		/* No disk memory. So we can not convert last unformatted node
		   to the direct item.  In this case we used to adjust
		   indirect items's ih_free_space. Now ih_free_space is not
		   used, it would be ideal to write zeros to corresponding
		   unformatted node. For now i_size is considered as guard for
		   going out of file size */
		kunmap(page);
J
Jeff Mahoney 已提交
260
		return block_size - round_tail_len;
261 262
	}
	kunmap(page);
L
Linus Torvalds 已提交
263

264
	/* make sure to get the i_blocks changes from reiserfs_insert_item */
265
	reiserfs_update_sd(th, inode);
L
Linus Torvalds 已提交
266

267 268 269
	// note: we have now the same as in above direct2indirect
	// conversion: there are two keys which have matching first three
	// key components. They only differ by the fouhth one.
L
Linus Torvalds 已提交
270

271 272
	/* We have inserted new direct item and must remove last
	   unformatted node. */
J
Jeff Mahoney 已提交
273
	*mode = M_CUT;
L
Linus Torvalds 已提交
274

275
	/* we store position of first direct item in the in-core inode */
276 277
	/* mark_file_with_tail (inode, pos1 + 1); */
	REISERFS_I(inode)->i_first_direct_byte = pos1 + 1;
L
Linus Torvalds 已提交
278

J
Jeff Mahoney 已提交
279
	return block_size - round_tail_len;
280
}