blob: 66bdaa2cf496adefcc4a9940a36da544491dada0 [file] [log] [blame]
Ryusuke Konishiae980432018-09-04 15:46:30 -07001// SPDX-License-Identifier: GPL-2.0+
Ryusuke Konishia60be982009-04-06 19:01:25 -07002/*
Ryusuke Konishi94ee1d92021-11-08 18:35:01 -08003 * NILFS B-tree node cache
Ryusuke Konishia60be982009-04-06 19:01:25 -07004 *
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 *
Ryusuke Konishi4b420ab2016-05-23 16:23:09 -07007 * Originally written by Seiji Kihara.
8 * Fully revised by Ryusuke Konishi for stabilization and simplification.
Ryusuke Konishia60be982009-04-06 19:01:25 -07009 *
10 */
11
12#include <linux/types.h>
13#include <linux/buffer_head.h>
14#include <linux/mm.h>
15#include <linux/backing-dev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090016#include <linux/gfp.h>
Ryusuke Konishia60be982009-04-06 19:01:25 -070017#include "nilfs.h"
18#include "mdt.h"
19#include "dat.h"
20#include "page.h"
21#include "btnode.h"
22
Ryusuke Konishia60be982009-04-06 19:01:25 -070023void nilfs_btnode_cache_clear(struct address_space *btnc)
24{
25 invalidate_mapping_pages(btnc, 0, -1);
26 truncate_inode_pages(btnc, 0);
27}
28
Ryusuke Konishid501d732009-11-13 16:04:11 +090029struct buffer_head *
30nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
31{
32 struct inode *inode = NILFS_BTNC_I(btnc);
33 struct buffer_head *bh;
34
Ryusuke Konishi4ce5c342016-08-02 14:05:28 -070035 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
Ryusuke Konishid501d732009-11-13 16:04:11 +090036 if (unlikely(!bh))
37 return NULL;
38
39 if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
40 buffer_dirty(bh))) {
41 brelse(bh);
42 BUG();
43 }
Fabian Frederick93407472017-02-27 14:28:32 -080044 memset(bh->b_data, 0, i_blocksize(inode));
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +090045 bh->b_bdev = inode->i_sb->s_bdev;
Ryusuke Konishid501d732009-11-13 16:04:11 +090046 bh->b_blocknr = blocknr;
47 set_buffer_mapped(bh);
48 set_buffer_uptodate(bh);
49
50 unlock_page(bh->b_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030051 put_page(bh->b_page);
Ryusuke Konishid501d732009-11-13 16:04:11 +090052 return bh;
53}
54
Ryusuke Konishia60be982009-04-06 19:01:25 -070055int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
Mike Christie2a222ca2016-06-05 14:31:43 -050056 sector_t pblocknr, int mode, int mode_flags,
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +090057 struct buffer_head **pbh, sector_t *submit_ptr)
Ryusuke Konishia60be982009-04-06 19:01:25 -070058{
59 struct buffer_head *bh;
60 struct inode *inode = NILFS_BTNC_I(btnc);
Ryusuke Konishif8e6cc02010-07-15 11:39:10 +090061 struct page *page;
Ryusuke Konishia60be982009-04-06 19:01:25 -070062 int err;
63
Ryusuke Konishi4ce5c342016-08-02 14:05:28 -070064 bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
Ryusuke Konishia60be982009-04-06 19:01:25 -070065 if (unlikely(!bh))
66 return -ENOMEM;
67
68 err = -EEXIST; /* internal code */
Ryusuke Konishif8e6cc02010-07-15 11:39:10 +090069 page = bh->b_page;
Ryusuke Konishia60be982009-04-06 19:01:25 -070070
71 if (buffer_uptodate(bh) || buffer_dirty(bh))
72 goto found;
73
74 if (pblocknr == 0) {
75 pblocknr = blocknr;
76 if (inode->i_ino != NILFS_DAT_INO) {
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +090077 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
Ryusuke Konishia60be982009-04-06 19:01:25 -070078
79 /* blocknr is a virtual block number */
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +090080 err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
81 &pblocknr);
Ryusuke Konishia60be982009-04-06 19:01:25 -070082 if (unlikely(err)) {
83 brelse(bh);
84 goto out_locked;
85 }
86 }
87 }
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +090088
Mike Christie2a222ca2016-06-05 14:31:43 -050089 if (mode_flags & REQ_RAHEAD) {
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +090090 if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
91 err = -EBUSY; /* internal code */
92 brelse(bh);
93 goto out_locked;
94 }
95 } else { /* mode == READ */
96 lock_buffer(bh);
97 }
Ryusuke Konishia60be982009-04-06 19:01:25 -070098 if (buffer_uptodate(bh)) {
99 unlock_buffer(bh);
100 err = -EEXIST; /* internal code */
101 goto found;
102 }
103 set_buffer_mapped(bh);
Ryusuke Konishi0ef28f92011-05-05 12:56:51 +0900104 bh->b_bdev = inode->i_sb->s_bdev;
Ryusuke Konishia60be982009-04-06 19:01:25 -0700105 bh->b_blocknr = pblocknr; /* set block address for read */
106 bh->b_end_io = end_buffer_read_sync;
107 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -0500108 submit_bh(mode, mode_flags, bh);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700109 bh->b_blocknr = blocknr; /* set back to the given block address */
Ryusuke Konishi26dfdd82010-07-18 10:42:23 +0900110 *submit_ptr = pblocknr;
Ryusuke Konishia60be982009-04-06 19:01:25 -0700111 err = 0;
112found:
113 *pbh = bh;
114
115out_locked:
Ryusuke Konishif8e6cc02010-07-15 11:39:10 +0900116 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300117 put_page(page);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700118 return err;
119}
120
Ryusuke Konishia60be982009-04-06 19:01:25 -0700121/**
122 * nilfs_btnode_delete - delete B-tree node buffer
123 * @bh: buffer to be deleted
124 *
125 * nilfs_btnode_delete() invalidates the specified buffer and delete the page
126 * including the buffer if the page gets unbusy.
127 */
128void nilfs_btnode_delete(struct buffer_head *bh)
129{
130 struct address_space *mapping;
131 struct page *page = bh->b_page;
132 pgoff_t index = page_index(page);
133 int still_dirty;
134
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300135 get_page(page);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700136 lock_page(page);
137 wait_on_page_writeback(page);
138
139 nilfs_forget_buffer(bh);
140 still_dirty = PageDirty(page);
141 mapping = page->mapping;
142 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300143 put_page(page);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700144
145 if (!still_dirty && mapping)
146 invalidate_inode_pages2_range(mapping, index, index);
147}
148
149/**
150 * nilfs_btnode_prepare_change_key
151 * prepare to move contents of the block for old key to one of new key.
152 * the old buffer will not be removed, but might be reused for new buffer.
153 * it might return -ENOMEM because of memory allocation errors,
154 * and might return -EIO because of disk read errors.
155 */
156int nilfs_btnode_prepare_change_key(struct address_space *btnc,
157 struct nilfs_btnode_chkey_ctxt *ctxt)
158{
159 struct buffer_head *obh, *nbh;
160 struct inode *inode = NILFS_BTNC_I(btnc);
161 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
162 int err;
163
164 if (oldkey == newkey)
165 return 0;
166
167 obh = ctxt->bh;
168 ctxt->newbh = NULL;
169
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300170 if (inode->i_blkbits == PAGE_SHIFT) {
Matthew Wilcoxf611ff62017-12-04 19:33:30 -0500171 struct page *opage = obh->b_page;
172 lock_page(opage);
Ryusuke Konishib1f1b8c2009-08-30 04:21:41 +0900173retry:
Ryusuke Konishia60be982009-04-06 19:01:25 -0700174 /* BUG_ON(oldkey != obh->b_page->index); */
Matthew Wilcoxf611ff62017-12-04 19:33:30 -0500175 if (unlikely(oldkey != opage->index))
176 NILFS_PAGE_BUG(opage,
Ryusuke Konishia60be982009-04-06 19:01:25 -0700177 "invalid oldkey %lld (newkey=%lld)",
178 (unsigned long long)oldkey,
179 (unsigned long long)newkey);
180
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700181 xa_lock_irq(&btnc->i_pages);
Matthew Wilcoxf611ff62017-12-04 19:33:30 -0500182 err = __xa_insert(&btnc->i_pages, newkey, opage, GFP_NOFS);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700183 xa_unlock_irq(&btnc->i_pages);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700184 /*
185 * Note: page->index will not change to newkey until
186 * nilfs_btnode_commit_change_key() will be called.
187 * To protect the page in intermediate state, the page lock
188 * is held.
189 */
Ryusuke Konishia60be982009-04-06 19:01:25 -0700190 if (!err)
191 return 0;
Matthew Wilcoxfd9dc932019-02-06 13:07:11 -0500192 else if (err != -EBUSY)
Ryusuke Konishia60be982009-04-06 19:01:25 -0700193 goto failed_unlock;
194
195 err = invalidate_inode_pages2_range(btnc, newkey, newkey);
196 if (!err)
197 goto retry;
198 /* fallback to copy mode */
Matthew Wilcoxf611ff62017-12-04 19:33:30 -0500199 unlock_page(opage);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700200 }
201
Ryusuke Konishi45f49102009-11-13 16:25:19 +0900202 nbh = nilfs_btnode_create_block(btnc, newkey);
203 if (!nbh)
204 return -ENOMEM;
205
206 BUG_ON(nbh == obh);
207 ctxt->newbh = nbh;
208 return 0;
Ryusuke Konishia60be982009-04-06 19:01:25 -0700209
210 failed_unlock:
211 unlock_page(obh->b_page);
212 return err;
213}
214
215/**
216 * nilfs_btnode_commit_change_key
217 * commit the change_key operation prepared by prepare_change_key().
218 */
219void nilfs_btnode_commit_change_key(struct address_space *btnc,
220 struct nilfs_btnode_chkey_ctxt *ctxt)
221{
222 struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
223 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
224 struct page *opage;
225
226 if (oldkey == newkey)
227 return;
228
229 if (nbh == NULL) { /* blocksize == pagesize */
230 opage = obh->b_page;
231 if (unlikely(oldkey != opage->index))
232 NILFS_PAGE_BUG(opage,
233 "invalid oldkey %lld (newkey=%lld)",
234 (unsigned long long)oldkey,
235 (unsigned long long)newkey);
Ryusuke Konishi5fc7b142011-05-05 12:56:51 +0900236 mark_buffer_dirty(obh);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700237
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700238 xa_lock_irq(&btnc->i_pages);
Matthew Wilcoxf611ff62017-12-04 19:33:30 -0500239 __xa_erase(&btnc->i_pages, oldkey);
240 __xa_set_mark(&btnc->i_pages, newkey, PAGECACHE_TAG_DIRTY);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700241 xa_unlock_irq(&btnc->i_pages);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700242
243 opage->index = obh->b_blocknr = newkey;
244 unlock_page(opage);
245 } else {
246 nilfs_copy_buffer(nbh, obh);
Ryusuke Konishi5fc7b142011-05-05 12:56:51 +0900247 mark_buffer_dirty(nbh);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700248
249 nbh->b_blocknr = newkey;
250 ctxt->bh = nbh;
251 nilfs_btnode_delete(obh); /* will decrement bh->b_count */
252 }
253}
254
255/**
256 * nilfs_btnode_abort_change_key
257 * abort the change_key operation prepared by prepare_change_key().
258 */
259void nilfs_btnode_abort_change_key(struct address_space *btnc,
260 struct nilfs_btnode_chkey_ctxt *ctxt)
261{
262 struct buffer_head *nbh = ctxt->newbh;
263 __u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
264
265 if (oldkey == newkey)
266 return;
267
268 if (nbh == NULL) { /* blocksize == pagesize */
Matthew Wilcoxfe2b5112017-12-04 19:33:30 -0500269 xa_erase_irq(&btnc->i_pages, newkey);
Ryusuke Konishia60be982009-04-06 19:01:25 -0700270 unlock_page(ctxt->bh->b_page);
271 } else
272 brelse(nbh);
273}