blob: c21e0b4454a6762a2d6bbbed84dd03858afa2185 [file] [log] [blame]
Ryusuke Konishia60be982009-04-06 19:01:25 -07001/*
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 *
Ryusuke Konishi4b420ab2016-05-23 16:23:09 -070016 * Originally written by Seiji Kihara.
17 * Fully revised by Ryusuke Konishi for stabilization and simplification.
Ryusuke Konishia60be982009-04-06 19:01:25 -070018 *
19 */
20
21#include <linux/types.h>
22#include <linux/buffer_head.h>
23#include <linux/mm.h>
24#include <linux/backing-dev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/gfp.h>
Ryusuke Konishia60be982009-04-06 19:01:25 -070026#include "nilfs.h"
27#include "mdt.h"
28#include "dat.h"
29#include "page.h"
30#include "btnode.h"
31
Ryusuke Konishia60be982009-04-06 19:01:25 -070032void nilfs_btnode_cache_clear(struct address_space *btnc)
33{
34 invalidate_mapping_pages(btnc, 0, -1);
35 truncate_inode_pages(btnc, 0);
36}
37
Ryusuke Konishid501d732009-11-13 16:04:11 +090038struct 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
Ryusuke Konishi4ce5c342016-08-02 14:05:28 -070044 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
Ryusuke Konishid501d732009-11-13 16:04:11 +090045 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 }
Fabian Frederick93407472017-02-27 14:28:32 -080053 memset(bh->b_data, 0, i_blocksize(inode));
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +090054 bh->b_bdev = inode->i_sb->s_bdev;
Ryusuke Konishid501d732009-11-13 16:04:11 +090055 bh->b_blocknr = blocknr;
56 set_buffer_mapped(bh);
57 set_buffer_uptodate(bh);
58
59 unlock_page(bh->b_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030060 put_page(bh->b_page);
Ryusuke Konishid501d732009-11-13 16:04:11 +090061 return bh;
62}
63
Ryusuke Konishia60be982009-04-06 19:01:25 -070064int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
Mike Christie2a222ca2016-06-05 14:31:43 -050065 sector_t pblocknr, int mode, int mode_flags,
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +090066 struct buffer_head **pbh, sector_t *submit_ptr)
Ryusuke Konishia60be982009-04-06 19:01:25 -070067{
68 struct buffer_head *bh;
69 struct inode *inode = NILFS_BTNC_I(btnc);
Ryusuke Konishif8e6cc02010-07-15 11:39:10 +090070 struct page *page;
Ryusuke Konishia60be982009-04-06 19:01:25 -070071 int err;
72
Ryusuke Konishi4ce5c342016-08-02 14:05:28 -070073 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
Ryusuke Konishia60be982009-04-06 19:01:25 -070074 if (unlikely(!bh))
75 return -ENOMEM;
76
77 err = -EEXIST; /* internal code */
Ryusuke Konishif8e6cc02010-07-15 11:39:10 +090078 page = bh->b_page;
Ryusuke Konishia60be982009-04-06 19:01:25 -070079
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) {
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +090086 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
Ryusuke Konishia60be982009-04-06 19:01:25 -070087
88 /* blocknr is a virtual block number */
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +090089 err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
90 &pblocknr);
Ryusuke Konishia60be982009-04-06 19:01:25 -070091 if (unlikely(err)) {
92 brelse(bh);
93 goto out_locked;
94 }
95 }
96 }
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +090097
Mike Christie2a222ca2016-06-05 14:31:43 -050098 if (mode_flags & REQ_RAHEAD) {
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +090099 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 }
Ryusuke Konishia60be982009-04-06 19:01:25 -0700107 if (buffer_uptodate(bh)) {
108 unlock_buffer(bh);
109 err = -EEXIST; /* internal code */
110 goto found;
111 }
112 set_buffer_mapped(bh);
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +0900113 bh->b_bdev = inode->i_sb->s_bdev;
Ryusuke Konishia60be982009-04-06 19:01:25 -0700114 bh->b_blocknr = pblocknr; /* set block address for read */
115 bh->b_end_io = end_buffer_read_sync;
116 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -0500117 submit_bh(mode, mode_flags, bh);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700118 bh->b_blocknr = blocknr; /* set back to the given block address */
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +0900119 *submit_ptr = pblocknr;
Ryusuke Konishia60be982009-04-06 19:01:25 -0700120 err = 0;
121found:
122 *pbh = bh;
123
124out_locked:
Ryusuke Konishif8e6cc02010-07-15 11:39:10 +0900125 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300126 put_page(page);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700127 return err;
128}
129
Ryusuke Konishia60be982009-04-06 19:01:25 -0700130/**
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
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300144 get_page(page);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700145 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);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300152 put_page(page);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700153
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
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300179 if (inode->i_blkbits == PAGE_SHIFT) {
Ryusuke Konishia60be982009-04-06 19:01:25 -0700180 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 */
Ryusuke Konishib1f1b8c2009-08-30 04:21:41 +0900185retry:
Ryusuke Konishia60be982009-04-06 19:01:25 -0700186 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
Ryusuke Konishia60be982009-04-06 19:01:25 -0700196 spin_lock_irq(&btnc->tree_lock);
197 err = radix_tree_insert(&btnc->page_tree, newkey, obh->b_page);
198 spin_unlock_irq(&btnc->tree_lock);
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
Ryusuke Konishi45f49102009-11-13 16:25:19 +0900218 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;
Ryusuke Konishia60be982009-04-06 19:01:25 -0700225
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);
Ryusuke Konishi5fc7b142011-05-05 12:56:51 +0900252 mark_buffer_dirty(obh);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700253
254 spin_lock_irq(&btnc->tree_lock);
255 radix_tree_delete(&btnc->page_tree, oldkey);
256 radix_tree_tag_set(&btnc->page_tree, newkey,
257 PAGECACHE_TAG_DIRTY);
258 spin_unlock_irq(&btnc->tree_lock);
259
260 opage->index = obh->b_blocknr = newkey;
261 unlock_page(opage);
262 } else {
263 nilfs_copy_buffer(nbh, obh);
Ryusuke Konishi5fc7b142011-05-05 12:56:51 +0900264 mark_buffer_dirty(nbh);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700265
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 spin_lock_irq(&btnc->tree_lock);
287 radix_tree_delete(&btnc->page_tree, newkey);
288 spin_unlock_irq(&btnc->tree_lock);
289 unlock_page(ctxt->bh->b_page);
290 } else
291 brelse(nbh);
292}