Loading...
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 * Originally written by Seiji Kihara.
17 * Fully revised by Ryusuke Konishi for stabilization and simplification.
18 *
19 */
20
21#include <linux/types.h>
22#include <linux/buffer_head.h>
23#include <linux/mm.h>
24#include <linux/backing-dev.h>
25#include <linux/gfp.h>
26#include "nilfs.h"
27#include "mdt.h"
28#include "dat.h"
29#include "page.h"
30#include "btnode.h"
31
32void nilfs_btnode_cache_clear(struct address_space *btnc)
33{
34 invalidate_mapping_pages(btnc, 0, -1);
35 truncate_inode_pages(btnc, 0);
36}
37
38struct buffer_head *
39nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
40{
41 struct inode *inode = NILFS_BTNC_I(btnc);
42 struct buffer_head *bh;
43
44 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
45 if (unlikely(!bh))
46 return NULL;
47
48 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
49 buffer_dirty(bh))) {
50 brelse(bh);
51 BUG();
52 }
53 memset(bh->b_data, 0, i_blocksize(inode));
54 bh->b_bdev = inode->i_sb->s_bdev;
55 bh->b_blocknr = blocknr;
56 set_buffer_mapped(bh);
57 set_buffer_uptodate(bh);
58
59 unlock_page(bh->b_page);
60 put_page(bh->b_page);
61 return bh;
62}
63
64int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
65 sector_t pblocknr, int mode, int mode_flags,
66 struct buffer_head **pbh, sector_t *submit_ptr)
67{
68 struct buffer_head *bh;
69 struct inode *inode = NILFS_BTNC_I(btnc);
70 struct page *page;
71 int err;
72
73 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
74 if (unlikely(!bh))
75 return -ENOMEM;
76
77 err = -EEXIST; /* internal code */
78 page = bh->b_page;
79
80 if (buffer_uptodate(bh) || buffer_dirty(bh))
81 goto found;
82
83 if (pblocknr == 0) {
84 pblocknr = blocknr;
85 if (inode->i_ino != NILFS_DAT_INO) {
86 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
87
88 /* blocknr is a virtual block number */
89 err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
90 &pblocknr);
91 if (unlikely(err)) {
92 brelse(bh);
93 goto out_locked;
94 }
95 }
96 }
97
98 if (mode_flags & REQ_RAHEAD) {
99 if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
100 err = -EBUSY; /* internal code */
101 brelse(bh);
102 goto out_locked;
103 }
104 } else { /* mode == READ */
105 lock_buffer(bh);
106 }
107 if (buffer_uptodate(bh)) {
108 unlock_buffer(bh);
109 err = -EEXIST; /* internal code */
110 goto found;
111 }
112 set_buffer_mapped(bh);
113 bh->b_bdev = inode->i_sb->s_bdev;
114 bh->b_blocknr = pblocknr; /* set block address for read */
115 bh->b_end_io = end_buffer_read_sync;
116 get_bh(bh);
117 submit_bh(mode, mode_flags, bh);
118 bh->b_blocknr = blocknr; /* set back to the given block address */
119 *submit_ptr = pblocknr;
120 err = 0;
121found:
122 *pbh = bh;
123
124out_locked:
125 unlock_page(page);
126 put_page(page);
127 return err;
128}
129
130/**
131 * nilfs_btnode_delete - delete B-tree node buffer
132 * @bh: buffer to be deleted
133 *
134 * nilfs_btnode_delete() invalidates the specified buffer and delete the page
135 * including the buffer if the page gets unbusy.
136 */
137void nilfs_btnode_delete(struct buffer_head *bh)
138{
139 struct address_space *mapping;
140 struct page *page = bh->b_page;
141 pgoff_t index = page_index(page);
142 int still_dirty;
143
144 get_page(page);
145 lock_page(page);
146 wait_on_page_writeback(page);
147
148 nilfs_forget_buffer(bh);
149 still_dirty = PageDirty(page);
150 mapping = page->mapping;
151 unlock_page(page);
152 put_page(page);
153
154 if (!still_dirty && mapping)
155 invalidate_inode_pages2_range(mapping, index, index);
156}
157
158/**
159 * nilfs_btnode_prepare_change_key
160 * prepare to move contents of the block for old key to one of new key.
161 * the old buffer will not be removed, but might be reused for new buffer.
162 * it might return -ENOMEM because of memory allocation errors,
163 * and might return -EIO because of disk read errors.
164 */
165int nilfs_btnode_prepare_change_key(struct address_space *btnc,
166 struct nilfs_btnode_chkey_ctxt *ctxt)
167{
168 struct buffer_head *obh, *nbh;
169 struct inode *inode = NILFS_BTNC_I(btnc);
170 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
171 int err;
172
173 if (oldkey == newkey)
174 return 0;
175
176 obh = ctxt->bh;
177 ctxt->newbh = NULL;
178
179 if (inode->i_blkbits == PAGE_SHIFT) {
180 lock_page(obh->b_page);
181 /*
182 * We cannot call radix_tree_preload for the kernels older
183 * than 2.6.23, because it is not exported for modules.
184 */
185retry:
186 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
187 if (err)
188 goto failed_unlock;
189 /* BUG_ON(oldkey != obh->b_page->index); */
190 if (unlikely(oldkey != obh->b_page->index))
191 NILFS_PAGE_BUG(obh->b_page,
192 "invalid oldkey %lld (newkey=%lld)",
193 (unsigned long long)oldkey,
194 (unsigned long long)newkey);
195
196 xa_lock_irq(&btnc->i_pages);
197 err = radix_tree_insert(&btnc->i_pages, newkey, obh->b_page);
198 xa_unlock_irq(&btnc->i_pages);
199 /*
200 * Note: page->index will not change to newkey until
201 * nilfs_btnode_commit_change_key() will be called.
202 * To protect the page in intermediate state, the page lock
203 * is held.
204 */
205 radix_tree_preload_end();
206 if (!err)
207 return 0;
208 else if (err != -EEXIST)
209 goto failed_unlock;
210
211 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
212 if (!err)
213 goto retry;
214 /* fallback to copy mode */
215 unlock_page(obh->b_page);
216 }
217
218 nbh = nilfs_btnode_create_block(btnc, newkey);
219 if (!nbh)
220 return -ENOMEM;
221
222 BUG_ON(nbh == obh);
223 ctxt->newbh = nbh;
224 return 0;
225
226 failed_unlock:
227 unlock_page(obh->b_page);
228 return err;
229}
230
231/**
232 * nilfs_btnode_commit_change_key
233 * commit the change_key operation prepared by prepare_change_key().
234 */
235void nilfs_btnode_commit_change_key(struct address_space *btnc,
236 struct nilfs_btnode_chkey_ctxt *ctxt)
237{
238 struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
239 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
240 struct page *opage;
241
242 if (oldkey == newkey)
243 return;
244
245 if (nbh == NULL) { /* blocksize == pagesize */
246 opage = obh->b_page;
247 if (unlikely(oldkey != opage->index))
248 NILFS_PAGE_BUG(opage,
249 "invalid oldkey %lld (newkey=%lld)",
250 (unsigned long long)oldkey,
251 (unsigned long long)newkey);
252 mark_buffer_dirty(obh);
253
254 xa_lock_irq(&btnc->i_pages);
255 radix_tree_delete(&btnc->i_pages, oldkey);
256 radix_tree_tag_set(&btnc->i_pages, newkey,
257 PAGECACHE_TAG_DIRTY);
258 xa_unlock_irq(&btnc->i_pages);
259
260 opage->index = obh->b_blocknr = newkey;
261 unlock_page(opage);
262 } else {
263 nilfs_copy_buffer(nbh, obh);
264 mark_buffer_dirty(nbh);
265
266 nbh->b_blocknr = newkey;
267 ctxt->bh = nbh;
268 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
269 }
270}
271
272/**
273 * nilfs_btnode_abort_change_key
274 * abort the change_key operation prepared by prepare_change_key().
275 */
276void nilfs_btnode_abort_change_key(struct address_space *btnc,
277 struct nilfs_btnode_chkey_ctxt *ctxt)
278{
279 struct buffer_head *nbh = ctxt->newbh;
280 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
281
282 if (oldkey == newkey)
283 return;
284
285 if (nbh == NULL) { /* blocksize == pagesize */
286 xa_lock_irq(&btnc->i_pages);
287 radix_tree_delete(&btnc->i_pages, newkey);
288 xa_unlock_irq(&btnc->i_pages);
289 unlock_page(ctxt->bh->b_page);
290 } else
291 brelse(nbh);
292}
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
37void nilfs_btnode_cache_clear(struct address_space *btnc)
38{
39 invalidate_mapping_pages(btnc, 0, -1);
40 truncate_inode_pages(btnc, 0);
41}
42
43struct buffer_head *
44nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
45{
46 struct inode *inode = NILFS_BTNC_I(btnc);
47 struct buffer_head *bh;
48
49 bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
50 if (unlikely(!bh))
51 return NULL;
52
53 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
54 buffer_dirty(bh))) {
55 brelse(bh);
56 BUG();
57 }
58 memset(bh->b_data, 0, 1 << inode->i_blkbits);
59 bh->b_bdev = inode->i_sb->s_bdev;
60 bh->b_blocknr = blocknr;
61 set_buffer_mapped(bh);
62 set_buffer_uptodate(bh);
63
64 unlock_page(bh->b_page);
65 page_cache_release(bh->b_page);
66 return bh;
67}
68
69int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
70 sector_t pblocknr, int mode,
71 struct buffer_head **pbh, sector_t *submit_ptr)
72{
73 struct buffer_head *bh;
74 struct inode *inode = NILFS_BTNC_I(btnc);
75 struct page *page;
76 int err;
77
78 bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node);
79 if (unlikely(!bh))
80 return -ENOMEM;
81
82 err = -EEXIST; /* internal code */
83 page = bh->b_page;
84
85 if (buffer_uptodate(bh) || buffer_dirty(bh))
86 goto found;
87
88 if (pblocknr == 0) {
89 pblocknr = blocknr;
90 if (inode->i_ino != NILFS_DAT_INO) {
91 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
92
93 /* blocknr is a virtual block number */
94 err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
95 &pblocknr);
96 if (unlikely(err)) {
97 brelse(bh);
98 goto out_locked;
99 }
100 }
101 }
102
103 if (mode == READA) {
104 if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
105 err = -EBUSY; /* internal code */
106 brelse(bh);
107 goto out_locked;
108 }
109 } else { /* mode == READ */
110 lock_buffer(bh);
111 }
112 if (buffer_uptodate(bh)) {
113 unlock_buffer(bh);
114 err = -EEXIST; /* internal code */
115 goto found;
116 }
117 set_buffer_mapped(bh);
118 bh->b_bdev = inode->i_sb->s_bdev;
119 bh->b_blocknr = pblocknr; /* set block address for read */
120 bh->b_end_io = end_buffer_read_sync;
121 get_bh(bh);
122 submit_bh(mode, bh);
123 bh->b_blocknr = blocknr; /* set back to the given block address */
124 *submit_ptr = pblocknr;
125 err = 0;
126found:
127 *pbh = bh;
128
129out_locked:
130 unlock_page(page);
131 page_cache_release(page);
132 return err;
133}
134
135/**
136 * nilfs_btnode_delete - delete B-tree node buffer
137 * @bh: buffer to be deleted
138 *
139 * nilfs_btnode_delete() invalidates the specified buffer and delete the page
140 * including the buffer if the page gets unbusy.
141 */
142void nilfs_btnode_delete(struct buffer_head *bh)
143{
144 struct address_space *mapping;
145 struct page *page = bh->b_page;
146 pgoff_t index = page_index(page);
147 int still_dirty;
148
149 page_cache_get(page);
150 lock_page(page);
151 wait_on_page_writeback(page);
152
153 nilfs_forget_buffer(bh);
154 still_dirty = PageDirty(page);
155 mapping = page->mapping;
156 unlock_page(page);
157 page_cache_release(page);
158
159 if (!still_dirty && mapping)
160 invalidate_inode_pages2_range(mapping, index, index);
161}
162
163/**
164 * nilfs_btnode_prepare_change_key
165 * prepare to move contents of the block for old key to one of new key.
166 * the old buffer will not be removed, but might be reused for new buffer.
167 * it might return -ENOMEM because of memory allocation errors,
168 * and might return -EIO because of disk read errors.
169 */
170int nilfs_btnode_prepare_change_key(struct address_space *btnc,
171 struct nilfs_btnode_chkey_ctxt *ctxt)
172{
173 struct buffer_head *obh, *nbh;
174 struct inode *inode = NILFS_BTNC_I(btnc);
175 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
176 int err;
177
178 if (oldkey == newkey)
179 return 0;
180
181 obh = ctxt->bh;
182 ctxt->newbh = NULL;
183
184 if (inode->i_blkbits == PAGE_CACHE_SHIFT) {
185 lock_page(obh->b_page);
186 /*
187 * We cannot call radix_tree_preload for the kernels older
188 * than 2.6.23, because it is not exported for modules.
189 */
190retry:
191 err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
192 if (err)
193 goto failed_unlock;
194 /* BUG_ON(oldkey != obh->b_page->index); */
195 if (unlikely(oldkey != obh->b_page->index))
196 NILFS_PAGE_BUG(obh->b_page,
197 "invalid oldkey %lld (newkey=%lld)",
198 (unsigned long long)oldkey,
199 (unsigned long long)newkey);
200
201 spin_lock_irq(&btnc->tree_lock);
202 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
203 spin_unlock_irq(&btnc->tree_lock);
204 /*
205 * Note: page->index will not change to newkey until
206 * nilfs_btnode_commit_change_key() will be called.
207 * To protect the page in intermediate state, the page lock
208 * is held.
209 */
210 radix_tree_preload_end();
211 if (!err)
212 return 0;
213 else if (err != -EEXIST)
214 goto failed_unlock;
215
216 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
217 if (!err)
218 goto retry;
219 /* fallback to copy mode */
220 unlock_page(obh->b_page);
221 }
222
223 nbh = nilfs_btnode_create_block(btnc, newkey);
224 if (!nbh)
225 return -ENOMEM;
226
227 BUG_ON(nbh == obh);
228 ctxt->newbh = nbh;
229 return 0;
230
231 failed_unlock:
232 unlock_page(obh->b_page);
233 return err;
234}
235
236/**
237 * nilfs_btnode_commit_change_key
238 * commit the change_key operation prepared by prepare_change_key().
239 */
240void nilfs_btnode_commit_change_key(struct address_space *btnc,
241 struct nilfs_btnode_chkey_ctxt *ctxt)
242{
243 struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
244 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
245 struct page *opage;
246
247 if (oldkey == newkey)
248 return;
249
250 if (nbh == NULL) { /* blocksize == pagesize */
251 opage = obh->b_page;
252 if (unlikely(oldkey != opage->index))
253 NILFS_PAGE_BUG(opage,
254 "invalid oldkey %lld (newkey=%lld)",
255 (unsigned long long)oldkey,
256 (unsigned long long)newkey);
257 mark_buffer_dirty(obh);
258
259 spin_lock_irq(&btnc->tree_lock);
260 radix_tree_delete(&btnc->page_tree, oldkey);
261 radix_tree_tag_set(&btnc->page_tree, newkey,
262 PAGECACHE_TAG_DIRTY);
263 spin_unlock_irq(&btnc->tree_lock);
264
265 opage->index = obh->b_blocknr = newkey;
266 unlock_page(opage);
267 } else {
268 nilfs_copy_buffer(nbh, obh);
269 mark_buffer_dirty(nbh);
270
271 nbh->b_blocknr = newkey;
272 ctxt->bh = nbh;
273 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
274 }
275}
276
277/**
278 * nilfs_btnode_abort_change_key
279 * abort the change_key operation prepared by prepare_change_key().
280 */
281void nilfs_btnode_abort_change_key(struct address_space *btnc,
282 struct nilfs_btnode_chkey_ctxt *ctxt)
283{
284 struct buffer_head *nbh = ctxt->newbh;
285 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
286
287 if (oldkey == newkey)
288 return;
289
290 if (nbh == NULL) { /* blocksize == pagesize */
291 spin_lock_irq(&btnc->tree_lock);
292 radix_tree_delete(&btnc->page_tree, newkey);
293 spin_unlock_irq(&btnc->tree_lock);
294 unlock_page(ctxt->bh->b_page);
295 } else
296 brelse(nbh);
297}