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 
nilfs_btnode_cache_init(struct address_space * btnc,struct backing_dev_info * bdi)37 void nilfs_btnode_cache_init(struct address_space *btnc,
38 			     struct backing_dev_info *bdi)
39 {
40 	nilfs_mapping_init(btnc, bdi);
41 }
42 
nilfs_btnode_cache_clear(struct address_space * btnc)43 void nilfs_btnode_cache_clear(struct address_space *btnc)
44 {
45 	invalidate_mapping_pages(btnc, 0, -1);
46 	truncate_inode_pages(btnc, 0);
47 }
48 
49 struct buffer_head *
nilfs_btnode_create_block(struct address_space * btnc,__u64 blocknr)50 nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
51 {
52 	struct inode *inode = NILFS_BTNC_I(btnc);
53 	struct buffer_head *bh;
54 
55 	bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
56 	if (unlikely(!bh))
57 		return NULL;
58 
59 	if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
60 		     buffer_dirty(bh))) {
61 		brelse(bh);
62 		BUG();
63 	}
64 	memset(bh->b_data, 0, 1 << inode->i_blkbits);
65 	bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
66 	bh->b_blocknr = blocknr;
67 	set_buffer_mapped(bh);
68 	set_buffer_uptodate(bh);
69 
70 	unlock_page(bh->b_page);
71 	page_cache_release(bh->b_page);
72 	return bh;
73 }
74 
nilfs_btnode_submit_block(struct address_space * btnc,__u64 blocknr,sector_t pblocknr,int mode,struct buffer_head ** pbh,sector_t * submit_ptr)75 int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
76 			      sector_t pblocknr, int mode,
77 			      struct buffer_head **pbh, sector_t *submit_ptr)
78 {
79 	struct buffer_head *bh;
80 	struct inode *inode = NILFS_BTNC_I(btnc);
81 	struct page *page;
82 	int err;
83 
84 	bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
85 	if (unlikely(!bh))
86 		return -ENOMEM;
87 
88 	err = -EEXIST; /* internal code */
89 	page = bh->b_page;
90 
91 	if (buffer_uptodate(bh) || buffer_dirty(bh))
92 		goto found;
93 
94 	if (pblocknr == 0) {
95 		pblocknr = blocknr;
96 		if (inode->i_ino != NILFS_DAT_INO) {
97 			struct inode *dat = NILFS_I_NILFS(inode)->ns_dat;
98 
99 			/* blocknr is a virtual block number */
100 			err = nilfs_dat_translate(dat, blocknr, &pblocknr);
101 			if (unlikely(err)) {
102 				brelse(bh);
103 				goto out_locked;
104 			}
105 		}
106 	}
107 
108 	if (mode == READA) {
109 		if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
110 			err = -EBUSY; /* internal code */
111 			brelse(bh);
112 			goto out_locked;
113 		}
114 	} else { /* mode == READ */
115 		lock_buffer(bh);
116 	}
117 	if (buffer_uptodate(bh)) {
118 		unlock_buffer(bh);
119 		err = -EEXIST; /* internal code */
120 		goto found;
121 	}
122 	set_buffer_mapped(bh);
123 	bh->b_bdev = NILFS_I_NILFS(inode)->ns_bdev;
124 	bh->b_blocknr = pblocknr; /* set block address for read */
125 	bh->b_end_io = end_buffer_read_sync;
126 	get_bh(bh);
127 	submit_bh(mode, bh);
128 	bh->b_blocknr = blocknr; /* set back to the given block address */
129 	*submit_ptr = pblocknr;
130 	err = 0;
131 found:
132 	*pbh = bh;
133 
134 out_locked:
135 	unlock_page(page);
136 	page_cache_release(page);
137 	return err;
138 }
139 
140 /**
141  * nilfs_btnode_delete - delete B-tree node buffer
142  * @bh: buffer to be deleted
143  *
144  * nilfs_btnode_delete() invalidates the specified buffer and delete the page
145  * including the buffer if the page gets unbusy.
146  */
nilfs_btnode_delete(struct buffer_head * bh)147 void nilfs_btnode_delete(struct buffer_head *bh)
148 {
149 	struct address_space *mapping;
150 	struct page *page = bh->b_page;
151 	pgoff_t index = page_index(page);
152 	int still_dirty;
153 
154 	page_cache_get(page);
155 	lock_page(page);
156 	wait_on_page_writeback(page);
157 
158 	nilfs_forget_buffer(bh);
159 	still_dirty = PageDirty(page);
160 	mapping = page->mapping;
161 	unlock_page(page);
162 	page_cache_release(page);
163 
164 	if (!still_dirty && mapping)
165 		invalidate_inode_pages2_range(mapping, index, index);
166 }
167 
168 /**
169  * nilfs_btnode_prepare_change_key
170  *  prepare to move contents of the block for old key to one of new key.
171  *  the old buffer will not be removed, but might be reused for new buffer.
172  *  it might return -ENOMEM because of memory allocation errors,
173  *  and might return -EIO because of disk read errors.
174  */
nilfs_btnode_prepare_change_key(struct address_space * btnc,struct nilfs_btnode_chkey_ctxt * ctxt)175 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
176 				    struct nilfs_btnode_chkey_ctxt *ctxt)
177 {
178 	struct buffer_head *obh, *nbh;
179 	struct inode *inode = NILFS_BTNC_I(btnc);
180 	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
181 	int err;
182 
183 	if (oldkey == newkey)
184 		return 0;
185 
186 	obh = ctxt->bh;
187 	ctxt->newbh = NULL;
188 
189 	if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
190 		lock_page(obh->b_page);
191 		/*
192 		 * We cannot call radix_tree_preload for the kernels older
193 		 * than 2.6.23, because it is not exported for modules.
194 		 */
195 retry:
196 		err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
197 		if (err)
198 			goto failed_unlock;
199 		/* BUG_ON(oldkey != obh->b_page->index); */
200 		if (unlikely(oldkey != obh->b_page->index))
201 			NILFS_PAGE_BUG(obh->b_page,
202 				       "invalid oldkey %lld (newkey=%lld)",
203 				       (unsigned long long)oldkey,
204 				       (unsigned long long)newkey);
205 
206 		spin_lock_irq(&btnc->tree_lock);
207 		err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
208 		spin_unlock_irq(&btnc->tree_lock);
209 		/*
210 		 * Note: page->index will not change to newkey until
211 		 * nilfs_btnode_commit_change_key() will be called.
212 		 * To protect the page in intermediate state, the page lock
213 		 * is held.
214 		 */
215 		radix_tree_preload_end();
216 		if (!err)
217 			return 0;
218 		else if (err != -EEXIST)
219 			goto failed_unlock;
220 
221 		err = invalidate_inode_pages2_range(btnc, newkey, newkey);
222 		if (!err)
223 			goto retry;
224 		/* fallback to copy mode */
225 		unlock_page(obh->b_page);
226 	}
227 
228 	nbh = nilfs_btnode_create_block(btnc, newkey);
229 	if (!nbh)
230 		return -ENOMEM;
231 
232 	BUG_ON(nbh == obh);
233 	ctxt->newbh = nbh;
234 	return 0;
235 
236  failed_unlock:
237 	unlock_page(obh->b_page);
238 	return err;
239 }
240 
241 /**
242  * nilfs_btnode_commit_change_key
243  *  commit the change_key operation prepared by prepare_change_key().
244  */
nilfs_btnode_commit_change_key(struct address_space * btnc,struct nilfs_btnode_chkey_ctxt * ctxt)245 void nilfs_btnode_commit_change_key(struct address_space *btnc,
246 				    struct nilfs_btnode_chkey_ctxt *ctxt)
247 {
248 	struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
249 	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
250 	struct page *opage;
251 
252 	if (oldkey == newkey)
253 		return;
254 
255 	if (nbh == NULL) {	/* blocksize == pagesize */
256 		opage = obh->b_page;
257 		if (unlikely(oldkey != opage->index))
258 			NILFS_PAGE_BUG(opage,
259 				       "invalid oldkey %lld (newkey=%lld)",
260 				       (unsigned long long)oldkey,
261 				       (unsigned long long)newkey);
262 		nilfs_btnode_mark_dirty(obh);
263 
264 		spin_lock_irq(&btnc->tree_lock);
265 		radix_tree_delete(&btnc->page_tree, oldkey);
266 		radix_tree_tag_set(&btnc->page_tree, newkey,
267 				   PAGECACHE_TAG_DIRTY);
268 		spin_unlock_irq(&btnc->tree_lock);
269 
270 		opage->index = obh->b_blocknr = newkey;
271 		unlock_page(opage);
272 	} else {
273 		nilfs_copy_buffer(nbh, obh);
274 		nilfs_btnode_mark_dirty(nbh);
275 
276 		nbh->b_blocknr = newkey;
277 		ctxt->bh = nbh;
278 		nilfs_btnode_delete(obh); /* will decrement bh->b_count */
279 	}
280 }
281 
282 /**
283  * nilfs_btnode_abort_change_key
284  *  abort the change_key operation prepared by prepare_change_key().
285  */
nilfs_btnode_abort_change_key(struct address_space * btnc,struct nilfs_btnode_chkey_ctxt * ctxt)286 void nilfs_btnode_abort_change_key(struct address_space *btnc,
287 				   struct nilfs_btnode_chkey_ctxt *ctxt)
288 {
289 	struct buffer_head *nbh = ctxt->newbh;
290 	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
291 
292 	if (oldkey == newkey)
293 		return;
294 
295 	if (nbh == NULL) {	/* blocksize == pagesize */
296 		spin_lock_irq(&btnc->tree_lock);
297 		radix_tree_delete(&btnc->page_tree, newkey);
298 		spin_unlock_irq(&btnc->tree_lock);
299 		unlock_page(ctxt->bh->b_page);
300 	} else
301 		brelse(nbh);
302 }
303