1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3 * NILFS B-tree node cache
4 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Originally written by Seiji Kihara.
8 * Fully revised by Ryusuke Konishi for stabilization and simplification.
9 *
10 */
11
12 #include <linux/types.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mm.h>
15 #include <linux/backing-dev.h>
16 #include <linux/gfp.h>
17 #include "nilfs.h"
18 #include "mdt.h"
19 #include "dat.h"
20 #include "page.h"
21 #include "btnode.h"
22
23
24 /**
25 * nilfs_init_btnc_inode - initialize B-tree node cache inode
26 * @btnc_inode: inode to be initialized
27 *
28 * nilfs_init_btnc_inode() sets up an inode for B-tree node cache.
29 */
30 void nilfs_init_btnc_inode(struct inode *btnc_inode)
31 {
32 struct nilfs_inode_info *ii = NILFS_I(btnc_inode);
33
34 btnc_inode->i_mode = S_IFREG;
35 ii->i_flags = 0;
36 memset(&ii->i_bmap_data, 0, sizeof(struct nilfs_bmap));
37 mapping_set_gfp_mask(btnc_inode->i_mapping, GFP_NOFS);
38 }
39
40 void nilfs_btnode_cache_clear(struct address_space *btnc)
41 {
42 invalidate_mapping_pages(btnc, 0, -1);
43 truncate_inode_pages(btnc, 0);
44 }
45
46 struct buffer_head *
47 nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
48 {
49 struct inode *inode = btnc->host;
50 struct buffer_head *bh;
51
52 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
53 if (unlikely(!bh))
54 return ERR_PTR(-ENOMEM);
55
56 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
57 buffer_dirty(bh))) {
58 /*
59 * The block buffer at the specified new address was already
60 * in use. This can happen if it is a virtual block number
61 * and has been reallocated due to corruption of the bitmap
62 * used to manage its allocation state (if not, the buffer
63 * clearing of an abandoned b-tree node is missing somewhere).
64 */
65 nilfs_error(inode->i_sb,
66 "state inconsistency probably due to duplicate use of b-tree node block address %llu (ino=%lu)",
67 (unsigned long long)blocknr, inode->i_ino);
68 goto failed;
69 }
70 memset(bh->b_data, 0, i_blocksize(inode));
71 bh->b_bdev = inode->i_sb->s_bdev;
72 bh->b_blocknr = blocknr;
73 set_buffer_mapped(bh);
74 set_buffer_uptodate(bh);
75
76 folio_unlock(bh->b_folio);
77 folio_put(bh->b_folio);
78 return bh;
79
80 failed:
81 folio_unlock(bh->b_folio);
82 folio_put(bh->b_folio);
83 brelse(bh);
84 return ERR_PTR(-EIO);
85 }
86
87 int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
88 sector_t pblocknr, blk_opf_t opf,
89 struct buffer_head **pbh, sector_t *submit_ptr)
90 {
91 struct buffer_head *bh;
92 struct inode *inode = btnc->host;
93 struct folio *folio;
94 int err;
95
96 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
97 if (unlikely(!bh))
98 return -ENOMEM;
99
100 err = -EEXIST; /* internal code */
101 folio = bh->b_folio;
102
103 if (buffer_uptodate(bh) || buffer_dirty(bh))
104 goto found;
105
106 if (pblocknr == 0) {
107 pblocknr = blocknr;
108 if (inode->i_ino != NILFS_DAT_INO) {
109 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
110
111 /* blocknr is a virtual block number */
112 err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
113 &pblocknr);
114 if (unlikely(err)) {
115 brelse(bh);
116 goto out_locked;
117 }
118 }
119 }
120
121 if (opf & REQ_RAHEAD) {
122 if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
123 err = -EBUSY; /* internal code */
124 brelse(bh);
125 goto out_locked;
126 }
127 } else { /* opf == REQ_OP_READ */
128 lock_buffer(bh);
129 }
130 if (buffer_uptodate(bh)) {
131 unlock_buffer(bh);
132 err = -EEXIST; /* internal code */
133 goto found;
134 }
135 set_buffer_mapped(bh);
136 bh->b_bdev = inode->i_sb->s_bdev;
137 bh->b_blocknr = pblocknr; /* set block address for read */
138 bh->b_end_io = end_buffer_read_sync;
139 get_bh(bh);
140 submit_bh(opf, bh);
141 bh->b_blocknr = blocknr; /* set back to the given block address */
142 *submit_ptr = pblocknr;
143 err = 0;
144 found:
145 *pbh = bh;
146
147 out_locked:
148 folio_unlock(folio);
149 folio_put(folio);
150 return err;
151 }
152
153 /**
154 * nilfs_btnode_delete - delete B-tree node buffer
155 * @bh: buffer to be deleted
156 *
157 * nilfs_btnode_delete() invalidates the specified buffer and delete the page
158 * including the buffer if the page gets unbusy.
159 */
160 void nilfs_btnode_delete(struct buffer_head *bh)
161 {
162 struct address_space *mapping;
163 struct folio *folio = bh->b_folio;
164 pgoff_t index = folio->index;
165 int still_dirty;
166
167 folio_get(folio);
168 folio_lock(folio);
169 folio_wait_writeback(folio);
170
171 nilfs_forget_buffer(bh);
172 still_dirty = folio_test_dirty(folio);
173 mapping = folio->mapping;
174 folio_unlock(folio);
175 folio_put(folio);
176
177 if (!still_dirty && mapping)
178 invalidate_inode_pages2_range(mapping, index, index);
179 }
180
181 /**
182 * nilfs_btnode_prepare_change_key
183 * prepare to move contents of the block for old key to one of new key.
184 * the old buffer will not be removed, but might be reused for new buffer.
185 * it might return -ENOMEM because of memory allocation errors,
186 * and might return -EIO because of disk read errors.
187 */
188 int nilfs_btnode_prepare_change_key(struct address_space *btnc,
189 struct nilfs_btnode_chkey_ctxt *ctxt)
190 {
191 struct buffer_head *obh, *nbh;
192 struct inode *inode = btnc->host;
193 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
194 int err;
195
196 if (oldkey == newkey)
197 return 0;
198
199 obh = ctxt->bh;
200 ctxt->newbh = NULL;
201
202 if (inode->i_blkbits == PAGE_SHIFT) {
203 struct folio *ofolio = obh->b_folio;
204 folio_lock(ofolio);
205 retry:
206 /* BUG_ON(oldkey != obh->b_folio->index); */
207 if (unlikely(oldkey != ofolio->index))
208 NILFS_FOLIO_BUG(ofolio,
209 "invalid oldkey %lld (newkey=%lld)",
210 (unsigned long long)oldkey,
211 (unsigned long long)newkey);
212
213 xa_lock_irq(&btnc->i_pages);
214 err = __xa_insert(&btnc->i_pages, newkey, ofolio, GFP_NOFS);
215 xa_unlock_irq(&btnc->i_pages);
216 /*
217 * Note: folio->index will not change to newkey until
218 * nilfs_btnode_commit_change_key() will be called.
219 * To protect the folio in intermediate state, the folio lock
220 * is held.
221 */
222 if (!err)
223 return 0;
224 else if (err != -EBUSY)
225 goto failed_unlock;
226
227 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
228 if (!err)
229 goto retry;
230 /* fallback to copy mode */
231 folio_unlock(ofolio);
232 }
233
234 nbh = nilfs_btnode_create_block(btnc, newkey);
235 if (IS_ERR(nbh))
236 return PTR_ERR(nbh);
237
238 BUG_ON(nbh == obh);
239 ctxt->newbh = nbh;
240 return 0;
241
242 failed_unlock:
243 folio_unlock(obh->b_folio);
244 return err;
245 }
246
247 /**
248 * nilfs_btnode_commit_change_key
249 * commit the change_key operation prepared by prepare_change_key().
250 */
251 void nilfs_btnode_commit_change_key(struct address_space *btnc,
252 struct nilfs_btnode_chkey_ctxt *ctxt)
253 {
254 struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
255 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
256 struct folio *ofolio;
257
258 if (oldkey == newkey)
259 return;
260
261 if (nbh == NULL) { /* blocksize == pagesize */
262 ofolio = obh->b_folio;
263 if (unlikely(oldkey != ofolio->index))
264 NILFS_FOLIO_BUG(ofolio,
265 "invalid oldkey %lld (newkey=%lld)",
266 (unsigned long long)oldkey,
267 (unsigned long long)newkey);
268 mark_buffer_dirty(obh);
269
270 xa_lock_irq(&btnc->i_pages);
271 __xa_erase(&btnc->i_pages, oldkey);
272 __xa_set_mark(&btnc->i_pages, newkey, PAGECACHE_TAG_DIRTY);
273 xa_unlock_irq(&btnc->i_pages);
274
275 ofolio->index = obh->b_blocknr = newkey;
276 folio_unlock(ofolio);
277 } else {
278 nilfs_copy_buffer(nbh, obh);
279 mark_buffer_dirty(nbh);
280
281 nbh->b_blocknr = newkey;
282 ctxt->bh = nbh;
283 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
284 }
285 }
286
287 /**
288 * nilfs_btnode_abort_change_key
289 * abort the change_key operation prepared by prepare_change_key().
290 */
291 void nilfs_btnode_abort_change_key(struct address_space *btnc,
292 struct nilfs_btnode_chkey_ctxt *ctxt)
293 {
294 struct buffer_head *nbh = ctxt->newbh;
295 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
296
297 if (oldkey == newkey)
298 return;
299
300 if (nbh == NULL) { /* blocksize == pagesize */
301 xa_erase_irq(&btnc->i_pages, newkey);
302 folio_unlock(ctxt->bh->b_folio);
303 } else {
304 /*
305 * When canceling a buffer that a prepare operation has
306 * allocated to copy a node block to another location, use
307 * nilfs_btnode_delete() to initialize and release the buffer
308 * so that the buffer flags will not be in an inconsistent
309 * state when it is reallocated.
310 */
311 nilfs_btnode_delete(nbh);
312 }
313 }
314
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.