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, true);
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 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
96 entry = kmap_local_folio(req->pr_entry_bh->b_folio, offset);
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);
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 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
120 entry = kmap_local_folio(req->pr_entry_bh->b_folio, offset);
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);
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 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
150 entry = kmap_local_folio(req->pr_entry_bh->b_folio, offset);
151 entry->de_start = cpu_to_le64(nilfs_mdt_cno(dat));
152 entry->de_blocknr = cpu_to_le64(blocknr);
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 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
172 entry = kmap_local_folio(req->pr_entry_bh->b_folio, offset);
173 start = le64_to_cpu(entry->de_start);
174 blocknr = le64_to_cpu(entry->de_blocknr);
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 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
207 entry = kmap_local_folio(req->pr_entry_bh->b_folio, offset);
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);
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 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
232 entry = kmap_local_folio(req->pr_entry_bh->b_folio, offset);
233 start = le64_to_cpu(entry->de_start);
234 blocknr = le64_to_cpu(entry->de_blocknr);
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 - mark the DAT block buffer containing the specified
275 * virtual block address entry as dirty
276 * @dat: DAT file inode
277 * @vblocknr: virtual block number
279 * Return: 0 on success, or the following negative error code on failure.
280 * * %-EINVAL - Invalid DAT entry (internal code).
281 * * %-EIO - I/O error (including metadata corruption).
282 * * %-ENOMEM - Insufficient memory available.
284 int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr)
286 struct nilfs_palloc_req req;
289 req.pr_entry_nr = vblocknr;
290 ret = nilfs_dat_prepare_entry(dat, &req, 0);
292 nilfs_dat_commit_entry(dat, &req);
297 * nilfs_dat_freev - free virtual block numbers
298 * @dat: DAT file inode
299 * @vblocknrs: array of virtual block numbers
300 * @nitems: number of virtual block numbers
302 * Description: nilfs_dat_freev() frees the virtual block numbers specified by
303 * @vblocknrs and @nitems.
305 * Return Value: On success, 0 is returned. On error, one of the following
306 * negative error codes is returned.
310 * %-ENOMEM - Insufficient amount of memory available.
312 * %-ENOENT - The virtual block number have not been allocated.
314 int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems)
316 return nilfs_palloc_freev(dat, vblocknrs, nitems);
320 * nilfs_dat_move - change a block number
321 * @dat: DAT file inode
322 * @vblocknr: virtual block number
323 * @blocknr: block number
325 * Description: nilfs_dat_move() changes the block number associated with
326 * @vblocknr to @blocknr.
328 * Return Value: On success, 0 is returned. On error, one of the following
329 * negative error codes is returned.
333 * %-ENOMEM - Insufficient amount of memory available.
335 int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
337 struct buffer_head *entry_bh;
338 struct nilfs_dat_entry *entry;
342 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
347 * The given disk block number (blocknr) is not yet written to
348 * the device at this point.
350 * To prevent nilfs_dat_translate() from returning the
351 * uncommitted block number, this makes a copy of the entry
352 * buffer and redirects nilfs_dat_translate() to the copy.
354 if (!buffer_nilfs_redirected(entry_bh)) {
355 ret = nilfs_mdt_freeze_buffer(dat, entry_bh);
362 offset = nilfs_palloc_entry_offset(dat, vblocknr, entry_bh);
363 entry = kmap_local_folio(entry_bh->b_folio, offset);
364 if (unlikely(entry->de_blocknr == cpu_to_le64(0))) {
365 nilfs_crit(dat->i_sb,
366 "%s: invalid vblocknr = %llu, [%llu, %llu)",
367 __func__, (unsigned long long)vblocknr,
368 (unsigned long long)le64_to_cpu(entry->de_start),
369 (unsigned long long)le64_to_cpu(entry->de_end));
374 WARN_ON(blocknr == 0);
375 entry->de_blocknr = cpu_to_le64(blocknr);
378 mark_buffer_dirty(entry_bh);
379 nilfs_mdt_mark_dirty(dat);
387 * nilfs_dat_translate - translate a virtual block number to a block number
388 * @dat: DAT file inode
389 * @vblocknr: virtual block number
390 * @blocknrp: pointer to a block number
392 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr
393 * to the corresponding block number.
395 * Return Value: On success, 0 is returned and the block number associated
396 * with @vblocknr is stored in the place pointed by @blocknrp. On error, one
397 * of the following negative error codes is returned.
401 * %-ENOMEM - Insufficient amount of memory available.
403 * %-ENOENT - A block number associated with @vblocknr does not exist.
405 int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
407 struct buffer_head *entry_bh, *bh;
408 struct nilfs_dat_entry *entry;
413 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
417 if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) {
418 bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh);
420 WARN_ON(!buffer_uptodate(bh));
426 offset = nilfs_palloc_entry_offset(dat, vblocknr, entry_bh);
427 entry = kmap_local_folio(entry_bh->b_folio, offset);
428 blocknr = le64_to_cpu(entry->de_blocknr);
441 ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz,
444 struct buffer_head *entry_bh;
445 struct nilfs_dat_entry *entry, *first_entry;
446 struct nilfs_vinfo *vinfo = buf;
449 unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block;
450 unsigned int entry_size = NILFS_MDT(dat)->mi_entry_size;
453 for (i = 0; i < nvi; i += n) {
454 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
459 first = vinfo->vi_vblocknr;
460 first = div64_ul(first, entries_per_block);
461 first *= entries_per_block;
462 /* first virtual block number in this block */
464 last = first + entries_per_block - 1;
465 /* last virtual block number in this block */
467 offset = nilfs_palloc_entry_offset(dat, first, entry_bh);
468 first_entry = kmap_local_folio(entry_bh->b_folio, offset);
470 j < nvi && vinfo->vi_vblocknr >= first &&
471 vinfo->vi_vblocknr <= last;
472 j++, n++, vinfo = (void *)vinfo + visz) {
473 entry = (void *)first_entry +
474 (vinfo->vi_vblocknr - first) * entry_size;
475 vinfo->vi_start = le64_to_cpu(entry->de_start);
476 vinfo->vi_end = le64_to_cpu(entry->de_end);
477 vinfo->vi_blocknr = le64_to_cpu(entry->de_blocknr);
479 kunmap_local(first_entry);
487 * nilfs_dat_read - read or get dat inode
488 * @sb: super block instance
489 * @entry_size: size of a dat entry
490 * @raw_inode: on-disk dat inode
491 * @inodep: buffer to store the inode
493 int nilfs_dat_read(struct super_block *sb, size_t entry_size,
494 struct nilfs_inode *raw_inode, struct inode **inodep)
496 static struct lock_class_key dat_lock_key;
498 struct nilfs_dat_info *di;
501 if (entry_size > sb->s_blocksize) {
502 nilfs_err(sb, "too large DAT entry size: %zu bytes",
505 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
506 nilfs_err(sb, "too small DAT entry size: %zu bytes",
511 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
514 if (!(dat->i_state & I_NEW))
517 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
521 err = nilfs_palloc_init_blockgroup(dat, entry_size);
525 di = NILFS_DAT_I(dat);
526 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
527 nilfs_palloc_setup_cache(dat, &di->palloc_cache);
528 err = nilfs_mdt_setup_shadow_map(dat, &di->shadow);
532 err = nilfs_read_inode_common(dat, raw_inode);
536 unlock_new_inode(dat);