Commit | Line | Data |
---|---|---|
a60be987 RK |
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 | * | |
a60be987 RK |
16 | * This file was originally written by Seiji Kihara <kihara@osrg.net> |
17 | * and fully revised by Ryusuke Konishi <ryusuke@osrg.net> for | |
18 | * stabilization and simplification. | |
19 | * | |
20 | */ | |
21 | ||
22 | #include <linux/types.h> | |
23 | #include <linux/buffer_head.h> | |
24 | #include <linux/mm.h> | |
25 | #include <linux/backing-dev.h> | |
5a0e3ad6 | 26 | #include <linux/gfp.h> |
a60be987 RK |
27 | #include "nilfs.h" |
28 | #include "mdt.h" | |
29 | #include "dat.h" | |
30 | #include "page.h" | |
31 | #include "btnode.h" | |
32 | ||
a60be987 RK |
33 | void nilfs_btnode_cache_clear(struct address_space *btnc) |
34 | { | |
35 | invalidate_mapping_pages(btnc, 0, -1); | |
36 | truncate_inode_pages(btnc, 0); | |
37 | } | |
38 | ||
d501d736 RK |
39 | struct buffer_head * |
40 | nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr) | |
41 | { | |
42 | struct inode *inode = NILFS_BTNC_I(btnc); | |
43 | struct buffer_head *bh; | |
44 | ||
45 | bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node); | |
46 | if (unlikely(!bh)) | |
47 | return NULL; | |
48 | ||
49 | if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) || | |
50 | buffer_dirty(bh))) { | |
51 | brelse(bh); | |
52 | BUG(); | |
53 | } | |
54 | memset(bh->b_data, 0, 1 << inode->i_blkbits); | |
0ef28f9a | 55 | bh->b_bdev = inode->i_sb->s_bdev; |
d501d736 RK |
56 | bh->b_blocknr = blocknr; |
57 | set_buffer_mapped(bh); | |
58 | set_buffer_uptodate(bh); | |
59 | ||
60 | unlock_page(bh->b_page); | |
09cbfeaf | 61 | put_page(bh->b_page); |
d501d736 RK |
62 | return bh; |
63 | } | |
64 | ||
a60be987 | 65 | int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr, |
26dfdd8e RK |
66 | sector_t pblocknr, int mode, |
67 | struct buffer_head **pbh, sector_t *submit_ptr) | |
a60be987 RK |
68 | { |
69 | struct buffer_head *bh; | |
70 | struct inode *inode = NILFS_BTNC_I(btnc); | |
f8e6cc01 | 71 | struct page *page; |
a60be987 RK |
72 | int err; |
73 | ||
74 | bh = nilfs_grab_buffer(inode, btnc, blocknr, 1 << BH_NILFS_Node); | |
75 | if (unlikely(!bh)) | |
76 | return -ENOMEM; | |
77 | ||
78 | err = -EEXIST; /* internal code */ | |
f8e6cc01 | 79 | page = bh->b_page; |
a60be987 RK |
80 | |
81 | if (buffer_uptodate(bh) || buffer_dirty(bh)) | |
82 | goto found; | |
83 | ||
84 | if (pblocknr == 0) { | |
85 | pblocknr = blocknr; | |
86 | if (inode->i_ino != NILFS_DAT_INO) { | |
0ef28f9a | 87 | struct the_nilfs *nilfs = inode->i_sb->s_fs_info; |
a60be987 RK |
88 | |
89 | /* blocknr is a virtual block number */ | |
0ef28f9a RK |
90 | err = nilfs_dat_translate(nilfs->ns_dat, blocknr, |
91 | &pblocknr); | |
a60be987 RK |
92 | if (unlikely(err)) { |
93 | brelse(bh); | |
94 | goto out_locked; | |
95 | } | |
96 | } | |
97 | } | |
26dfdd8e RK |
98 | |
99 | if (mode == READA) { | |
100 | if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) { | |
101 | err = -EBUSY; /* internal code */ | |
102 | brelse(bh); | |
103 | goto out_locked; | |
104 | } | |
105 | } else { /* mode == READ */ | |
106 | lock_buffer(bh); | |
107 | } | |
a60be987 RK |
108 | if (buffer_uptodate(bh)) { |
109 | unlock_buffer(bh); | |
110 | err = -EEXIST; /* internal code */ | |
111 | goto found; | |
112 | } | |
113 | set_buffer_mapped(bh); | |
0ef28f9a | 114 | bh->b_bdev = inode->i_sb->s_bdev; |
a60be987 RK |
115 | bh->b_blocknr = pblocknr; /* set block address for read */ |
116 | bh->b_end_io = end_buffer_read_sync; | |
117 | get_bh(bh); | |
26dfdd8e | 118 | submit_bh(mode, bh); |
a60be987 | 119 | bh->b_blocknr = blocknr; /* set back to the given block address */ |
26dfdd8e | 120 | *submit_ptr = pblocknr; |
a60be987 RK |
121 | err = 0; |
122 | found: | |
123 | *pbh = bh; | |
124 | ||
125 | out_locked: | |
f8e6cc01 | 126 | unlock_page(page); |
09cbfeaf | 127 | put_page(page); |
a60be987 RK |
128 | return err; |
129 | } | |
130 | ||
a60be987 RK |
131 | /** |
132 | * nilfs_btnode_delete - delete B-tree node buffer | |
133 | * @bh: buffer to be deleted | |
134 | * | |
135 | * nilfs_btnode_delete() invalidates the specified buffer and delete the page | |
136 | * including the buffer if the page gets unbusy. | |
137 | */ | |
138 | void nilfs_btnode_delete(struct buffer_head *bh) | |
139 | { | |
140 | struct address_space *mapping; | |
141 | struct page *page = bh->b_page; | |
142 | pgoff_t index = page_index(page); | |
143 | int still_dirty; | |
144 | ||
09cbfeaf | 145 | get_page(page); |
a60be987 RK |
146 | lock_page(page); |
147 | wait_on_page_writeback(page); | |
148 | ||
149 | nilfs_forget_buffer(bh); | |
150 | still_dirty = PageDirty(page); | |
151 | mapping = page->mapping; | |
152 | unlock_page(page); | |
09cbfeaf | 153 | put_page(page); |
a60be987 RK |
154 | |
155 | if (!still_dirty && mapping) | |
156 | invalidate_inode_pages2_range(mapping, index, index); | |
157 | } | |
158 | ||
159 | /** | |
160 | * nilfs_btnode_prepare_change_key | |
161 | * prepare to move contents of the block for old key to one of new key. | |
162 | * the old buffer will not be removed, but might be reused for new buffer. | |
163 | * it might return -ENOMEM because of memory allocation errors, | |
164 | * and might return -EIO because of disk read errors. | |
165 | */ | |
166 | int nilfs_btnode_prepare_change_key(struct address_space *btnc, | |
167 | struct nilfs_btnode_chkey_ctxt *ctxt) | |
168 | { | |
169 | struct buffer_head *obh, *nbh; | |
170 | struct inode *inode = NILFS_BTNC_I(btnc); | |
171 | __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey; | |
172 | int err; | |
173 | ||
174 | if (oldkey == newkey) | |
175 | return 0; | |
176 | ||
177 | obh = ctxt->bh; | |
178 | ctxt->newbh = NULL; | |
179 | ||
09cbfeaf | 180 | if (inode->i_blkbits == PAGE_SHIFT) { |
a60be987 RK |
181 | lock_page(obh->b_page); |
182 | /* | |
183 | * We cannot call radix_tree_preload for the kernels older | |
184 | * than 2.6.23, because it is not exported for modules. | |
185 | */ | |
b1f1b8ce | 186 | retry: |
a60be987 RK |
187 | err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); |
188 | if (err) | |
189 | goto failed_unlock; | |
190 | /* BUG_ON(oldkey != obh->b_page->index); */ | |
191 | if (unlikely(oldkey != obh->b_page->index)) | |
192 | NILFS_PAGE_BUG(obh->b_page, | |
193 | "invalid oldkey %lld (newkey=%lld)", | |
194 | (unsigned long long)oldkey, | |
195 | (unsigned long long)newkey); | |
196 | ||
a60be987 RK |
197 | spin_lock_irq(&btnc->tree_lock); |
198 | err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page); | |
199 | spin_unlock_irq(&btnc->tree_lock); | |
200 | /* | |
201 | * Note: page->index will not change to newkey until | |
202 | * nilfs_btnode_commit_change_key() will be called. | |
203 | * To protect the page in intermediate state, the page lock | |
204 | * is held. | |
205 | */ | |
206 | radix_tree_preload_end(); | |
207 | if (!err) | |
208 | return 0; | |
209 | else if (err != -EEXIST) | |
210 | goto failed_unlock; | |
211 | ||
212 | err = invalidate_inode_pages2_range(btnc, newkey, newkey); | |
213 | if (!err) | |
214 | goto retry; | |
215 | /* fallback to copy mode */ | |
216 | unlock_page(obh->b_page); | |
217 | } | |
218 | ||
45f4910b RK |
219 | nbh = nilfs_btnode_create_block(btnc, newkey); |
220 | if (!nbh) | |
221 | return -ENOMEM; | |
222 | ||
223 | BUG_ON(nbh == obh); | |
224 | ctxt->newbh = nbh; | |
225 | return 0; | |
a60be987 RK |
226 | |
227 | failed_unlock: | |
228 | unlock_page(obh->b_page); | |
229 | return err; | |
230 | } | |
231 | ||
232 | /** | |
233 | * nilfs_btnode_commit_change_key | |
234 | * commit the change_key operation prepared by prepare_change_key(). | |
235 | */ | |
236 | void nilfs_btnode_commit_change_key(struct address_space *btnc, | |
237 | struct nilfs_btnode_chkey_ctxt *ctxt) | |
238 | { | |
239 | struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh; | |
240 | __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey; | |
241 | struct page *opage; | |
242 | ||
243 | if (oldkey == newkey) | |
244 | return; | |
245 | ||
246 | if (nbh == NULL) { /* blocksize == pagesize */ | |
247 | opage = obh->b_page; | |
248 | if (unlikely(oldkey != opage->index)) | |
249 | NILFS_PAGE_BUG(opage, | |
250 | "invalid oldkey %lld (newkey=%lld)", | |
251 | (unsigned long long)oldkey, | |
252 | (unsigned long long)newkey); | |
5fc7b141 | 253 | mark_buffer_dirty(obh); |
a60be987 RK |
254 | |
255 | spin_lock_irq(&btnc->tree_lock); | |
256 | radix_tree_delete(&btnc->page_tree, oldkey); | |
257 | radix_tree_tag_set(&btnc->page_tree, newkey, | |
258 | PAGECACHE_TAG_DIRTY); | |
259 | spin_unlock_irq(&btnc->tree_lock); | |
260 | ||
261 | opage->index = obh->b_blocknr = newkey; | |
262 | unlock_page(opage); | |
263 | } else { | |
264 | nilfs_copy_buffer(nbh, obh); | |
5fc7b141 | 265 | mark_buffer_dirty(nbh); |
a60be987 RK |
266 | |
267 | nbh->b_blocknr = newkey; | |
268 | ctxt->bh = nbh; | |
269 | nilfs_btnode_delete(obh); /* will decrement bh->b_count */ | |
270 | } | |
271 | } | |
272 | ||
273 | /** | |
274 | * nilfs_btnode_abort_change_key | |
275 | * abort the change_key operation prepared by prepare_change_key(). | |
276 | */ | |
277 | void nilfs_btnode_abort_change_key(struct address_space *btnc, | |
278 | struct nilfs_btnode_chkey_ctxt *ctxt) | |
279 | { | |
280 | struct buffer_head *nbh = ctxt->newbh; | |
281 | __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey; | |
282 | ||
283 | if (oldkey == newkey) | |
284 | return; | |
285 | ||
286 | if (nbh == NULL) { /* blocksize == pagesize */ | |
287 | spin_lock_irq(&btnc->tree_lock); | |
288 | radix_tree_delete(&btnc->page_tree, newkey); | |
289 | spin_unlock_irq(&btnc->tree_lock); | |
290 | unlock_page(ctxt->bh->b_page); | |
291 | } else | |
292 | brelse(nbh); | |
293 | } |