xref: /openbmc/linux/fs/nilfs2/btnode.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /*
2  * btnode.c - NILFS B-tree node cache
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * This file was originally written by Seiji Kihara <kihara@osrg.net>
21  * and fully revised by Ryusuke Konishi <ryusuke@osrg.net> for
22  * stabilization and simplification.
23  *
24  */
25 
26 #include <linux/types.h>
27 #include <linux/buffer_head.h>
28 #include <linux/mm.h>
29 #include <linux/backing-dev.h>
30 #include <linux/gfp.h>
31 #include "nilfs.h"
32 #include "mdt.h"
33 #include "dat.h"
34 #include "page.h"
35 #include "btnode.h"
36 
37 
38 void nilfs_btnode_cache_init_once(struct address_space *btnc)
39 {
40 	nilfs_mapping_init_once(btnc);
41 }
42 
43 static const struct address_space_operations def_btnode_aops = {
44 	.sync_page		= block_sync_page,
45 };
46 
47 void nilfs_btnode_cache_init(struct address_space *btnc,
48 			     struct backing_dev_info *bdi)
49 {
50 	nilfs_mapping_init(btnc, bdi, &def_btnode_aops);
51 }
52 
53 void nilfs_btnode_cache_clear(struct address_space *btnc)
54 {
55 	invalidate_mapping_pages(btnc, 0, -1);
56 	truncate_inode_pages(btnc, 0);
57 }
58 
59 struct buffer_head *
60 nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
61 {
62 	struct inode *inode = NILFS_BTNC_I(btnc);
63 	struct buffer_head *bh;
64 
65 	bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
66 	if (unlikely(!bh))
67 		return NULL;
68 
69 	if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
70 		     buffer_dirty(bh))) {
71 		brelse(bh);
72 		BUG();
73 	}
74 	memset(bh->b_data, 0, 1 << inode->i_blkbits);
75 	bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
76 	bh->b_blocknr = blocknr;
77 	set_buffer_mapped(bh);
78 	set_buffer_uptodate(bh);
79 
80 	unlock_page(bh->b_page);
81 	page_cache_release(bh->b_page);
82 	return bh;
83 }
84 
85 int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
86 			      sector_t pblocknr, int mode,
87 			      struct buffer_head **pbh, sector_t *submit_ptr)
88 {
89 	struct buffer_head *bh;
90 	struct inode *inode = NILFS_BTNC_I(btnc);
91 	struct page *page;
92 	int err;
93 
94 	bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
95 	if (unlikely(!bh))
96 		return -ENOMEM;
97 
98 	err = -EEXIST; /* internal code */
99 	page = bh->b_page;
100 
101 	if (buffer_uptodate(bh) || buffer_dirty(bh))
102 		goto found;
103 
104 	if (pblocknr == 0) {
105 		pblocknr = blocknr;
106 		if (inode->i_ino != NILFS_DAT_INO) {
107 			struct inode *dat = NILFS_I_NILFS(inode)->ns_dat;
108 
109 			/* blocknr is a virtual block number */
110 			err = nilfs_dat_translate(dat, blocknr, &pblocknr);
111 			if (unlikely(err)) {
112 				brelse(bh);
113 				goto out_locked;
114 			}
115 		}
116 	}
117 
118 	if (mode == READA) {
119 		if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
120 			err = -EBUSY; /* internal code */
121 			brelse(bh);
122 			goto out_locked;
123 		}
124 	} else { /* mode == READ */
125 		lock_buffer(bh);
126 	}
127 	if (buffer_uptodate(bh)) {
128 		unlock_buffer(bh);
129 		err = -EEXIST; /* internal code */
130 		goto found;
131 	}
132 	set_buffer_mapped(bh);
133 	bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
134 	bh->b_blocknr = pblocknr; /* set block address for read */
135 	bh->b_end_io = end_buffer_read_sync;
136 	get_bh(bh);
137 	submit_bh(mode, bh);
138 	bh->b_blocknr = blocknr; /* set back to the given block address */
139 	*submit_ptr = pblocknr;
140 	err = 0;
141 found:
142 	*pbh = bh;
143 
144 out_locked:
145 	unlock_page(page);
146 	page_cache_release(page);
147 	return err;
148 }
149 
150 /**
151  * nilfs_btnode_delete - delete B-tree node buffer
152  * @bh: buffer to be deleted
153  *
154  * nilfs_btnode_delete() invalidates the specified buffer and delete the page
155  * including the buffer if the page gets unbusy.
156  */
157 void nilfs_btnode_delete(struct buffer_head *bh)
158 {
159 	struct address_space *mapping;
160 	struct page *page = bh->b_page;
161 	pgoff_t index = page_index(page);
162 	int still_dirty;
163 
164 	page_cache_get(page);
165 	lock_page(page);
166 	wait_on_page_writeback(page);
167 
168 	nilfs_forget_buffer(bh);
169 	still_dirty = PageDirty(page);
170 	mapping = page->mapping;
171 	unlock_page(page);
172 	page_cache_release(page);
173 
174 	if (!still_dirty && mapping)
175 		invalidate_inode_pages2_range(mapping, index, index);
176 }
177 
178 /**
179  * nilfs_btnode_prepare_change_key
180  *  prepare to move contents of the block for old key to one of new key.
181  *  the old buffer will not be removed, but might be reused for new buffer.
182  *  it might return -ENOMEM because of memory allocation errors,
183  *  and might return -EIO because of disk read errors.
184  */
185 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
186 				    struct nilfs_btnode_chkey_ctxt *ctxt)
187 {
188 	struct buffer_head *obh, *nbh;
189 	struct inode *inode = NILFS_BTNC_I(btnc);
190 	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
191 	int err;
192 
193 	if (oldkey == newkey)
194 		return 0;
195 
196 	obh = ctxt->bh;
197 	ctxt->newbh = NULL;
198 
199 	if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
200 		lock_page(obh->b_page);
201 		/*
202 		 * We cannot call radix_tree_preload for the kernels older
203 		 * than 2.6.23, because it is not exported for modules.
204 		 */
205 retry:
206 		err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
207 		if (err)
208 			goto failed_unlock;
209 		/* BUG_ON(oldkey != obh->b_page->index); */
210 		if (unlikely(oldkey != obh->b_page->index))
211 			NILFS_PAGE_BUG(obh->b_page,
212 				       "invalid oldkey %lld (newkey=%lld)",
213 				       (unsigned long long)oldkey,
214 				       (unsigned long long)newkey);
215 
216 		spin_lock_irq(&btnc->tree_lock);
217 		err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
218 		spin_unlock_irq(&btnc->tree_lock);
219 		/*
220 		 * Note: page->index will not change to newkey until
221 		 * nilfs_btnode_commit_change_key() will be called.
222 		 * To protect the page in intermediate state, the page lock
223 		 * is held.
224 		 */
225 		radix_tree_preload_end();
226 		if (!err)
227 			return 0;
228 		else if (err != -EEXIST)
229 			goto failed_unlock;
230 
231 		err = invalidate_inode_pages2_range(btnc, newkey, newkey);
232 		if (!err)
233 			goto retry;
234 		/* fallback to copy mode */
235 		unlock_page(obh->b_page);
236 	}
237 
238 	nbh = nilfs_btnode_create_block(btnc, newkey);
239 	if (!nbh)
240 		return -ENOMEM;
241 
242 	BUG_ON(nbh == obh);
243 	ctxt->newbh = nbh;
244 	return 0;
245 
246  failed_unlock:
247 	unlock_page(obh->b_page);
248 	return err;
249 }
250 
251 /**
252  * nilfs_btnode_commit_change_key
253  *  commit the change_key operation prepared by prepare_change_key().
254  */
255 void nilfs_btnode_commit_change_key(struct address_space *btnc,
256 				    struct nilfs_btnode_chkey_ctxt *ctxt)
257 {
258 	struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
259 	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
260 	struct page *opage;
261 
262 	if (oldkey == newkey)
263 		return;
264 
265 	if (nbh == NULL) {	/* blocksize == pagesize */
266 		opage = obh->b_page;
267 		if (unlikely(oldkey != opage->index))
268 			NILFS_PAGE_BUG(opage,
269 				       "invalid oldkey %lld (newkey=%lld)",
270 				       (unsigned long long)oldkey,
271 				       (unsigned long long)newkey);
272 		nilfs_btnode_mark_dirty(obh);
273 
274 		spin_lock_irq(&btnc->tree_lock);
275 		radix_tree_delete(&btnc->page_tree, oldkey);
276 		radix_tree_tag_set(&btnc->page_tree, newkey,
277 				   PAGECACHE_TAG_DIRTY);
278 		spin_unlock_irq(&btnc->tree_lock);
279 
280 		opage->index = obh->b_blocknr = newkey;
281 		unlock_page(opage);
282 	} else {
283 		nilfs_copy_buffer(nbh, obh);
284 		nilfs_btnode_mark_dirty(nbh);
285 
286 		nbh->b_blocknr = newkey;
287 		ctxt->bh = nbh;
288 		nilfs_btnode_delete(obh); /* will decrement bh->b_count */
289 	}
290 }
291 
292 /**
293  * nilfs_btnode_abort_change_key
294  *  abort the change_key operation prepared by prepare_change_key().
295  */
296 void nilfs_btnode_abort_change_key(struct address_space *btnc,
297 				   struct nilfs_btnode_chkey_ctxt *ctxt)
298 {
299 	struct buffer_head *nbh = ctxt->newbh;
300 	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
301 
302 	if (oldkey == newkey)
303 		return;
304 
305 	if (nbh == NULL) {	/* blocksize == pagesize */
306 		spin_lock_irq(&btnc->tree_lock);
307 		radix_tree_delete(&btnc->page_tree, newkey);
308 		spin_unlock_irq(&btnc->tree_lock);
309 		unlock_page(ctxt->bh->b_page);
310 	} else
311 		brelse(nbh);
312 }
313