1 // SPDX-License-Identifier: GPL-2.0+
3 * NILFS disk address translation.
5 * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Koji Sato.
10 #include <linux/types.h>
11 #include <linux/buffer_head.h>
12 #include <linux/string.h>
13 #include <linux/errno.h>
20 #define NILFS_CNO_MIN ((__u64)1)
21 #define NILFS_CNO_MAX (~(__u64)0)
24 * struct nilfs_dat_info - on-memory private data of DAT file
25 * @mi: on-memory private data of metadata file
26 * @palloc_cache: persistent object allocator cache of DAT file
27 * @shadow: shadow map of DAT file
29 struct nilfs_dat_info {
30 struct nilfs_mdt_info mi;
31 struct nilfs_palloc_cache palloc_cache;
32 struct nilfs_shadow_map shadow;
35 static inline struct nilfs_dat_info *NILFS_DAT_I(struct inode *dat)
37 return (struct nilfs_dat_info *)NILFS_MDT(dat);
40 static int nilfs_dat_prepare_entry(struct inode *dat,
41 struct nilfs_palloc_req *req, int create)
45 ret = nilfs_palloc_get_entry_block(dat, req->pr_entry_nr,
46 create, &req->pr_entry_bh);
47 if (unlikely(ret == -ENOENT)) {
49 "DAT doesn't have a block to manage vblocknr = %llu",
50 (unsigned long long)req->pr_entry_nr);
52 * Return internal code -EINVAL to notify bmap layer of
53 * metadata corruption.
60 static void nilfs_dat_commit_entry(struct inode *dat,
61 struct nilfs_palloc_req *req)
63 mark_buffer_dirty(req->pr_entry_bh);
64 nilfs_mdt_mark_dirty(dat);
65 brelse(req->pr_entry_bh);
68 static void nilfs_dat_abort_entry(struct inode *dat,
69 struct nilfs_palloc_req *req)
71 brelse(req->pr_entry_bh);
74 int nilfs_dat_prepare_alloc(struct inode *dat, struct nilfs_palloc_req *req)
78 ret = nilfs_palloc_prepare_alloc_entry(dat, req);
82 ret = nilfs_dat_prepare_entry(dat, req, 1);
84 nilfs_palloc_abort_alloc_entry(dat, req);
89 void nilfs_dat_commit_alloc(struct inode *dat, struct nilfs_palloc_req *req)
91 struct nilfs_dat_entry *entry;
94 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
95 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
96 req->pr_entry_bh, kaddr);
97 entry->de_start = cpu_to_le64(NILFS_CNO_MIN);
98 entry->de_end = cpu_to_le64(NILFS_CNO_MAX);
99 entry->de_blocknr = cpu_to_le64(0);
100 kunmap_atomic(kaddr);
102 nilfs_palloc_commit_alloc_entry(dat, req);
103 nilfs_dat_commit_entry(dat, req);
106 void nilfs_dat_abort_alloc(struct inode *dat, struct nilfs_palloc_req *req)
108 nilfs_dat_abort_entry(dat, req);
109 nilfs_palloc_abort_alloc_entry(dat, req);
112 static void nilfs_dat_commit_free(struct inode *dat,
113 struct nilfs_palloc_req *req)
115 struct nilfs_dat_entry *entry;
118 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
119 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
120 req->pr_entry_bh, kaddr);
121 entry->de_start = cpu_to_le64(NILFS_CNO_MIN);
122 entry->de_end = cpu_to_le64(NILFS_CNO_MIN);
123 entry->de_blocknr = cpu_to_le64(0);
124 kunmap_atomic(kaddr);
126 nilfs_dat_commit_entry(dat, req);
128 if (unlikely(req->pr_desc_bh == NULL || req->pr_bitmap_bh == NULL)) {
129 nilfs_error(dat->i_sb,
130 "state inconsistency probably due to duplicate use of vblocknr = %llu",
131 (unsigned long long)req->pr_entry_nr);
134 nilfs_palloc_commit_free_entry(dat, req);
137 int nilfs_dat_prepare_start(struct inode *dat, struct nilfs_palloc_req *req)
139 return nilfs_dat_prepare_entry(dat, req, 0);
142 void nilfs_dat_commit_start(struct inode *dat, struct nilfs_palloc_req *req,
145 struct nilfs_dat_entry *entry;
148 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
149 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
150 req->pr_entry_bh, kaddr);
151 entry->de_start = cpu_to_le64(nilfs_mdt_cno(dat));
152 entry->de_blocknr = cpu_to_le64(blocknr);
153 kunmap_atomic(kaddr);
155 nilfs_dat_commit_entry(dat, req);
158 int nilfs_dat_prepare_end(struct inode *dat, struct nilfs_palloc_req *req)
160 struct nilfs_dat_entry *entry;
166 ret = nilfs_dat_prepare_entry(dat, req, 0);
170 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
171 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
172 req->pr_entry_bh, kaddr);
173 start = le64_to_cpu(entry->de_start);
174 blocknr = le64_to_cpu(entry->de_blocknr);
175 kunmap_atomic(kaddr);
178 ret = nilfs_palloc_prepare_free_entry(dat, req);
180 nilfs_dat_abort_entry(dat, req);
184 if (unlikely(start > nilfs_mdt_cno(dat))) {
186 "vblocknr = %llu has abnormal lifetime: start cno (= %llu) > current cno (= %llu)",
187 (unsigned long long)req->pr_entry_nr,
188 (unsigned long long)start,
189 (unsigned long long)nilfs_mdt_cno(dat));
190 nilfs_dat_abort_entry(dat, req);
197 void nilfs_dat_commit_end(struct inode *dat, struct nilfs_palloc_req *req,
200 struct nilfs_dat_entry *entry;
205 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
206 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
207 req->pr_entry_bh, kaddr);
208 end = start = le64_to_cpu(entry->de_start);
210 end = nilfs_mdt_cno(dat);
211 WARN_ON(start > end);
213 entry->de_end = cpu_to_le64(end);
214 blocknr = le64_to_cpu(entry->de_blocknr);
215 kunmap_atomic(kaddr);
218 nilfs_dat_commit_free(dat, req);
220 nilfs_dat_commit_entry(dat, req);
223 void nilfs_dat_abort_end(struct inode *dat, struct nilfs_palloc_req *req)
225 struct nilfs_dat_entry *entry;
230 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
231 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
232 req->pr_entry_bh, kaddr);
233 start = le64_to_cpu(entry->de_start);
234 blocknr = le64_to_cpu(entry->de_blocknr);
235 kunmap_atomic(kaddr);
237 if (start == nilfs_mdt_cno(dat) && blocknr == 0)
238 nilfs_palloc_abort_free_entry(dat, req);
239 nilfs_dat_abort_entry(dat, req);
242 int nilfs_dat_prepare_update(struct inode *dat,
243 struct nilfs_palloc_req *oldreq,
244 struct nilfs_palloc_req *newreq)
248 ret = nilfs_dat_prepare_end(dat, oldreq);
250 ret = nilfs_dat_prepare_alloc(dat, newreq);
252 nilfs_dat_abort_end(dat, oldreq);
257 void nilfs_dat_commit_update(struct inode *dat,
258 struct nilfs_palloc_req *oldreq,
259 struct nilfs_palloc_req *newreq, int dead)
261 nilfs_dat_commit_end(dat, oldreq, dead);
262 nilfs_dat_commit_alloc(dat, newreq);
265 void nilfs_dat_abort_update(struct inode *dat,
266 struct nilfs_palloc_req *oldreq,
267 struct nilfs_palloc_req *newreq)
269 nilfs_dat_abort_end(dat, oldreq);
270 nilfs_dat_abort_alloc(dat, newreq);
274 * nilfs_dat_mark_dirty -
275 * @dat: DAT file inode
276 * @vblocknr: virtual block number
280 * Return Value: On success, 0 is returned. On error, one of the following
281 * negative error codes is returned.
285 * %-ENOMEM - Insufficient amount of memory available.
287 int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr)
289 struct nilfs_palloc_req req;
292 req.pr_entry_nr = vblocknr;
293 ret = nilfs_dat_prepare_entry(dat, &req, 0);
295 nilfs_dat_commit_entry(dat, &req);
300 * nilfs_dat_freev - free virtual block numbers
301 * @dat: DAT file inode
302 * @vblocknrs: array of virtual block numbers
303 * @nitems: number of virtual block numbers
305 * Description: nilfs_dat_freev() frees the virtual block numbers specified by
306 * @vblocknrs and @nitems.
308 * Return Value: On success, 0 is returned. On error, one of the following
309 * negative error codes is returned.
313 * %-ENOMEM - Insufficient amount of memory available.
315 * %-ENOENT - The virtual block number have not been allocated.
317 int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems)
319 return nilfs_palloc_freev(dat, vblocknrs, nitems);
323 * nilfs_dat_move - change a block number
324 * @dat: DAT file inode
325 * @vblocknr: virtual block number
326 * @blocknr: block number
328 * Description: nilfs_dat_move() changes the block number associated with
329 * @vblocknr to @blocknr.
331 * Return Value: On success, 0 is returned. On error, one of the following
332 * negative error codes is returned.
336 * %-ENOMEM - Insufficient amount of memory available.
338 int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
340 struct buffer_head *entry_bh;
341 struct nilfs_dat_entry *entry;
345 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
350 * The given disk block number (blocknr) is not yet written to
351 * the device at this point.
353 * To prevent nilfs_dat_translate() from returning the
354 * uncommitted block number, this makes a copy of the entry
355 * buffer and redirects nilfs_dat_translate() to the copy.
357 if (!buffer_nilfs_redirected(entry_bh)) {
358 ret = nilfs_mdt_freeze_buffer(dat, entry_bh);
365 kaddr = kmap_atomic(entry_bh->b_page);
366 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
367 if (unlikely(entry->de_blocknr == cpu_to_le64(0))) {
368 nilfs_crit(dat->i_sb,
369 "%s: invalid vblocknr = %llu, [%llu, %llu)",
370 __func__, (unsigned long long)vblocknr,
371 (unsigned long long)le64_to_cpu(entry->de_start),
372 (unsigned long long)le64_to_cpu(entry->de_end));
373 kunmap_atomic(kaddr);
377 WARN_ON(blocknr == 0);
378 entry->de_blocknr = cpu_to_le64(blocknr);
379 kunmap_atomic(kaddr);
381 mark_buffer_dirty(entry_bh);
382 nilfs_mdt_mark_dirty(dat);
390 * nilfs_dat_translate - translate a virtual block number to a block number
391 * @dat: DAT file inode
392 * @vblocknr: virtual block number
393 * @blocknrp: pointer to a block number
395 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr
396 * to the corresponding block number.
398 * Return Value: On success, 0 is returned and the block number associated
399 * with @vblocknr is stored in the place pointed by @blocknrp. On error, one
400 * of the following negative error codes is returned.
404 * %-ENOMEM - Insufficient amount of memory available.
406 * %-ENOENT - A block number associated with @vblocknr does not exist.
408 int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
410 struct buffer_head *entry_bh, *bh;
411 struct nilfs_dat_entry *entry;
416 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
420 if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) {
421 bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh);
423 WARN_ON(!buffer_uptodate(bh));
429 kaddr = kmap_atomic(entry_bh->b_page);
430 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
431 blocknr = le64_to_cpu(entry->de_blocknr);
439 kunmap_atomic(kaddr);
444 ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz,
447 struct buffer_head *entry_bh;
448 struct nilfs_dat_entry *entry;
449 struct nilfs_vinfo *vinfo = buf;
452 unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block;
455 for (i = 0; i < nvi; i += n) {
456 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
460 kaddr = kmap_atomic(entry_bh->b_page);
461 /* last virtual block number in this block */
462 first = vinfo->vi_vblocknr;
463 do_div(first, entries_per_block);
464 first *= entries_per_block;
465 last = first + entries_per_block - 1;
467 j < nvi && vinfo->vi_vblocknr >= first &&
468 vinfo->vi_vblocknr <= last;
469 j++, n++, vinfo = (void *)vinfo + visz) {
470 entry = nilfs_palloc_block_get_entry(
471 dat, vinfo->vi_vblocknr, entry_bh, kaddr);
472 vinfo->vi_start = le64_to_cpu(entry->de_start);
473 vinfo->vi_end = le64_to_cpu(entry->de_end);
474 vinfo->vi_blocknr = le64_to_cpu(entry->de_blocknr);
476 kunmap_atomic(kaddr);
484 * nilfs_dat_read - read or get dat inode
485 * @sb: super block instance
486 * @entry_size: size of a dat entry
487 * @raw_inode: on-disk dat inode
488 * @inodep: buffer to store the inode
490 int nilfs_dat_read(struct super_block *sb, size_t entry_size,
491 struct nilfs_inode *raw_inode, struct inode **inodep)
493 static struct lock_class_key dat_lock_key;
495 struct nilfs_dat_info *di;
498 if (entry_size > sb->s_blocksize) {
499 nilfs_err(sb, "too large DAT entry size: %zu bytes",
502 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
503 nilfs_err(sb, "too small DAT entry size: %zu bytes",
508 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
511 if (!(dat->i_state & I_NEW))
514 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
518 err = nilfs_palloc_init_blockgroup(dat, entry_size);
522 di = NILFS_DAT_I(dat);
523 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
524 nilfs_palloc_setup_cache(dat, &di->palloc_cache);
525 err = nilfs_mdt_setup_shadow_map(dat, &di->shadow);
529 err = nilfs_read_inode_common(dat, raw_inode);
533 unlock_new_inode(dat);