Loading...
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 */
30void 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 btnc_inode->i_mapping->a_ops = &nilfs_buffer_cache_aops;
39}
40
41void nilfs_btnode_cache_clear(struct address_space *btnc)
42{
43 invalidate_mapping_pages(btnc, 0, -1);
44 truncate_inode_pages(btnc, 0);
45}
46
47struct buffer_head *
48nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
49{
50 struct inode *inode = btnc->host;
51 struct buffer_head *bh;
52
53 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
54 if (unlikely(!bh))
55 return ERR_PTR(-ENOMEM);
56
57 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
58 buffer_dirty(bh))) {
59 /*
60 * The block buffer at the specified new address was already
61 * in use. This can happen if it is a virtual block number
62 * and has been reallocated due to corruption of the bitmap
63 * used to manage its allocation state (if not, the buffer
64 * clearing of an abandoned b-tree node is missing somewhere).
65 */
66 nilfs_error(inode->i_sb,
67 "state inconsistency probably due to duplicate use of b-tree node block address %llu (ino=%lu)",
68 (unsigned long long)blocknr, inode->i_ino);
69 goto failed;
70 }
71 memset(bh->b_data, 0, i_blocksize(inode));
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
80failed:
81 folio_unlock(bh->b_folio);
82 folio_put(bh->b_folio);
83 brelse(bh);
84 return ERR_PTR(-EIO);
85}
86
87int 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_blocknr = pblocknr; /* set block address for read */
137 bh->b_end_io = end_buffer_read_sync;
138 get_bh(bh);
139 submit_bh(opf, bh);
140 bh->b_blocknr = blocknr; /* set back to the given block address */
141 *submit_ptr = pblocknr;
142 err = 0;
143found:
144 *pbh = bh;
145
146out_locked:
147 folio_unlock(folio);
148 folio_put(folio);
149 return err;
150}
151
152/**
153 * nilfs_btnode_delete - delete B-tree node buffer
154 * @bh: buffer to be deleted
155 *
156 * nilfs_btnode_delete() invalidates the specified buffer and delete the page
157 * including the buffer if the page gets unbusy.
158 */
159void nilfs_btnode_delete(struct buffer_head *bh)
160{
161 struct address_space *mapping;
162 struct folio *folio = bh->b_folio;
163 pgoff_t index = folio->index;
164 int still_dirty;
165
166 folio_get(folio);
167 folio_lock(folio);
168 folio_wait_writeback(folio);
169
170 nilfs_forget_buffer(bh);
171 still_dirty = folio_test_dirty(folio);
172 mapping = folio->mapping;
173 folio_unlock(folio);
174 folio_put(folio);
175
176 if (!still_dirty && mapping)
177 invalidate_inode_pages2_range(mapping, index, index);
178}
179
180/**
181 * nilfs_btnode_prepare_change_key - prepare to change the search key of a
182 * b-tree node block
183 * @btnc: page cache in which the b-tree node block is buffered
184 * @ctxt: structure for exchanging context information for key change
185 *
186 * nilfs_btnode_prepare_change_key() prepares to move the contents of the
187 * b-tree node block of the old key given in the "oldkey" member of @ctxt to
188 * the position of the new key given in the "newkey" member of @ctxt in the
189 * page cache @btnc. Here, the key of the block is an index in units of
190 * blocks, and if the page and block sizes match, it matches the page index
191 * in the page cache.
192 *
193 * If the page size and block size match, this function attempts to move the
194 * entire folio, and in preparation for this, inserts the original folio into
195 * the new index of the cache. If this insertion fails or if the page size
196 * and block size are different, it falls back to a copy preparation using
197 * nilfs_btnode_create_block(), inserts a new block at the position
198 * corresponding to "newkey", and stores the buffer head pointer in the
199 * "newbh" member of @ctxt.
200 *
201 * Note that the current implementation does not support folio sizes larger
202 * than the page size.
203 *
204 * Return: 0 on success, or the following negative error code on failure.
205 * * %-EIO - I/O error (metadata corruption).
206 * * %-ENOMEM - Insufficient memory available.
207 */
208int nilfs_btnode_prepare_change_key(struct address_space *btnc,
209 struct nilfs_btnode_chkey_ctxt *ctxt)
210{
211 struct buffer_head *obh, *nbh;
212 struct inode *inode = btnc->host;
213 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
214 int err;
215
216 if (oldkey == newkey)
217 return 0;
218
219 obh = ctxt->bh;
220 ctxt->newbh = NULL;
221
222 if (inode->i_blkbits == PAGE_SHIFT) {
223 struct folio *ofolio = obh->b_folio;
224 folio_lock(ofolio);
225retry:
226 /* BUG_ON(oldkey != obh->b_folio->index); */
227 if (unlikely(oldkey != ofolio->index))
228 NILFS_FOLIO_BUG(ofolio,
229 "invalid oldkey %lld (newkey=%lld)",
230 (unsigned long long)oldkey,
231 (unsigned long long)newkey);
232
233 xa_lock_irq(&btnc->i_pages);
234 err = __xa_insert(&btnc->i_pages, newkey, ofolio, GFP_NOFS);
235 xa_unlock_irq(&btnc->i_pages);
236 /*
237 * Note: folio->index will not change to newkey until
238 * nilfs_btnode_commit_change_key() will be called.
239 * To protect the folio in intermediate state, the folio lock
240 * is held.
241 */
242 if (!err)
243 return 0;
244 else if (err != -EBUSY)
245 goto failed_unlock;
246
247 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
248 if (!err)
249 goto retry;
250 /* fallback to copy mode */
251 folio_unlock(ofolio);
252 }
253
254 nbh = nilfs_btnode_create_block(btnc, newkey);
255 if (IS_ERR(nbh))
256 return PTR_ERR(nbh);
257
258 BUG_ON(nbh == obh);
259 ctxt->newbh = nbh;
260 return 0;
261
262 failed_unlock:
263 folio_unlock(obh->b_folio);
264 return err;
265}
266
267/**
268 * nilfs_btnode_commit_change_key - commit the change of the search key of
269 * a b-tree node block
270 * @btnc: page cache in which the b-tree node block is buffered
271 * @ctxt: structure for exchanging context information for key change
272 *
273 * nilfs_btnode_commit_change_key() executes the key change based on the
274 * context @ctxt prepared by nilfs_btnode_prepare_change_key(). If no valid
275 * block buffer is prepared in "newbh" of @ctxt (i.e., a full folio move),
276 * this function removes the folio from the old index and completes the move.
277 * Otherwise, it copies the block data and inherited flag states of "oldbh"
278 * to "newbh" and clears the "oldbh" from the cache. In either case, the
279 * relocated buffer is marked as dirty.
280 *
281 * As with nilfs_btnode_prepare_change_key(), the current implementation does
282 * not support folio sizes larger than the page size.
283 */
284void nilfs_btnode_commit_change_key(struct address_space *btnc,
285 struct nilfs_btnode_chkey_ctxt *ctxt)
286{
287 struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
288 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
289 struct folio *ofolio;
290
291 if (oldkey == newkey)
292 return;
293
294 if (nbh == NULL) { /* blocksize == pagesize */
295 ofolio = obh->b_folio;
296 if (unlikely(oldkey != ofolio->index))
297 NILFS_FOLIO_BUG(ofolio,
298 "invalid oldkey %lld (newkey=%lld)",
299 (unsigned long long)oldkey,
300 (unsigned long long)newkey);
301 mark_buffer_dirty(obh);
302
303 xa_lock_irq(&btnc->i_pages);
304 __xa_erase(&btnc->i_pages, oldkey);
305 __xa_set_mark(&btnc->i_pages, newkey, PAGECACHE_TAG_DIRTY);
306 xa_unlock_irq(&btnc->i_pages);
307
308 ofolio->index = obh->b_blocknr = newkey;
309 folio_unlock(ofolio);
310 } else {
311 nilfs_copy_buffer(nbh, obh);
312 mark_buffer_dirty(nbh);
313
314 nbh->b_blocknr = newkey;
315 ctxt->bh = nbh;
316 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
317 }
318}
319
320/**
321 * nilfs_btnode_abort_change_key - abort the change of the search key of a
322 * b-tree node block
323 * @btnc: page cache in which the b-tree node block is buffered
324 * @ctxt: structure for exchanging context information for key change
325 *
326 * nilfs_btnode_abort_change_key() cancels the key change associated with the
327 * context @ctxt prepared via nilfs_btnode_prepare_change_key() and performs
328 * any necessary cleanup. If no valid block buffer is prepared in "newbh" of
329 * @ctxt, this function removes the folio from the destination index and aborts
330 * the move. Otherwise, it clears "newbh" from the cache.
331 *
332 * As with nilfs_btnode_prepare_change_key(), the current implementation does
333 * not support folio sizes larger than the page size.
334 */
335void nilfs_btnode_abort_change_key(struct address_space *btnc,
336 struct nilfs_btnode_chkey_ctxt *ctxt)
337{
338 struct buffer_head *nbh = ctxt->newbh;
339 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
340
341 if (oldkey == newkey)
342 return;
343
344 if (nbh == NULL) { /* blocksize == pagesize */
345 xa_erase_irq(&btnc->i_pages, newkey);
346 folio_unlock(ctxt->bh->b_folio);
347 } else {
348 /*
349 * When canceling a buffer that a prepare operation has
350 * allocated to copy a node block to another location, use
351 * nilfs_btnode_delete() to initialize and release the buffer
352 * so that the buffer flags will not be in an inconsistent
353 * state when it is reallocated.
354 */
355 nilfs_btnode_delete(nbh);
356 }
357}
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}