1 // SPDX-License-Identifier: GPL-2.0+
3 * dat.c - 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;
165 ret = nilfs_dat_prepare_entry(dat, req, 0);
169 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
170 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
171 req->pr_entry_bh, kaddr);
172 blocknr = le64_to_cpu(entry->de_blocknr);
173 kunmap_atomic(kaddr);
176 ret = nilfs_palloc_prepare_free_entry(dat, req);
178 nilfs_dat_abort_entry(dat, req);
186 void nilfs_dat_commit_end(struct inode *dat, struct nilfs_palloc_req *req,
189 struct nilfs_dat_entry *entry;
194 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
195 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
196 req->pr_entry_bh, kaddr);
197 end = start = le64_to_cpu(entry->de_start);
199 end = nilfs_mdt_cno(dat);
200 WARN_ON(start > end);
202 entry->de_end = cpu_to_le64(end);
203 blocknr = le64_to_cpu(entry->de_blocknr);
204 kunmap_atomic(kaddr);
207 nilfs_dat_commit_free(dat, req);
209 nilfs_dat_commit_entry(dat, req);
212 void nilfs_dat_abort_end(struct inode *dat, struct nilfs_palloc_req *req)
214 struct nilfs_dat_entry *entry;
219 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
220 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
221 req->pr_entry_bh, kaddr);
222 start = le64_to_cpu(entry->de_start);
223 blocknr = le64_to_cpu(entry->de_blocknr);
224 kunmap_atomic(kaddr);
226 if (start == nilfs_mdt_cno(dat) && blocknr == 0)
227 nilfs_palloc_abort_free_entry(dat, req);
228 nilfs_dat_abort_entry(dat, req);
231 int nilfs_dat_prepare_update(struct inode *dat,
232 struct nilfs_palloc_req *oldreq,
233 struct nilfs_palloc_req *newreq)
237 ret = nilfs_dat_prepare_end(dat, oldreq);
239 ret = nilfs_dat_prepare_alloc(dat, newreq);
241 nilfs_dat_abort_end(dat, oldreq);
246 void nilfs_dat_commit_update(struct inode *dat,
247 struct nilfs_palloc_req *oldreq,
248 struct nilfs_palloc_req *newreq, int dead)
250 nilfs_dat_commit_end(dat, oldreq, dead);
251 nilfs_dat_commit_alloc(dat, newreq);
254 void nilfs_dat_abort_update(struct inode *dat,
255 struct nilfs_palloc_req *oldreq,
256 struct nilfs_palloc_req *newreq)
258 nilfs_dat_abort_end(dat, oldreq);
259 nilfs_dat_abort_alloc(dat, newreq);
263 * nilfs_dat_mark_dirty -
264 * @dat: DAT file inode
265 * @vblocknr: virtual block number
269 * Return Value: On success, 0 is returned. On error, one of the following
270 * negative error codes is returned.
274 * %-ENOMEM - Insufficient amount of memory available.
276 int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr)
278 struct nilfs_palloc_req req;
281 req.pr_entry_nr = vblocknr;
282 ret = nilfs_dat_prepare_entry(dat, &req, 0);
284 nilfs_dat_commit_entry(dat, &req);
289 * nilfs_dat_freev - free virtual block numbers
290 * @dat: DAT file inode
291 * @vblocknrs: array of virtual block numbers
292 * @nitems: number of virtual block numbers
294 * Description: nilfs_dat_freev() frees the virtual block numbers specified by
295 * @vblocknrs and @nitems.
297 * Return Value: On success, 0 is returned. On error, one of the following
298 * negative error codes is returned.
302 * %-ENOMEM - Insufficient amount of memory available.
304 * %-ENOENT - The virtual block number have not been allocated.
306 int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems)
308 return nilfs_palloc_freev(dat, vblocknrs, nitems);
312 * nilfs_dat_move - change a block number
313 * @dat: DAT file inode
314 * @vblocknr: virtual block number
315 * @blocknr: block number
317 * Description: nilfs_dat_move() changes the block number associated with
318 * @vblocknr to @blocknr.
320 * Return Value: On success, 0 is returned. On error, one of the following
321 * negative error codes is returned.
325 * %-ENOMEM - Insufficient amount of memory available.
327 int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
329 struct buffer_head *entry_bh;
330 struct nilfs_dat_entry *entry;
334 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
339 * The given disk block number (blocknr) is not yet written to
340 * the device at this point.
342 * To prevent nilfs_dat_translate() from returning the
343 * uncommitted block number, this makes a copy of the entry
344 * buffer and redirects nilfs_dat_translate() to the copy.
346 if (!buffer_nilfs_redirected(entry_bh)) {
347 ret = nilfs_mdt_freeze_buffer(dat, entry_bh);
354 kaddr = kmap_atomic(entry_bh->b_page);
355 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
356 if (unlikely(entry->de_blocknr == cpu_to_le64(0))) {
357 nilfs_crit(dat->i_sb,
358 "%s: invalid vblocknr = %llu, [%llu, %llu)",
359 __func__, (unsigned long long)vblocknr,
360 (unsigned long long)le64_to_cpu(entry->de_start),
361 (unsigned long long)le64_to_cpu(entry->de_end));
362 kunmap_atomic(kaddr);
366 WARN_ON(blocknr == 0);
367 entry->de_blocknr = cpu_to_le64(blocknr);
368 kunmap_atomic(kaddr);
370 mark_buffer_dirty(entry_bh);
371 nilfs_mdt_mark_dirty(dat);
379 * nilfs_dat_translate - translate a virtual block number to a block number
380 * @dat: DAT file inode
381 * @vblocknr: virtual block number
382 * @blocknrp: pointer to a block number
384 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr
385 * to the corresponding block number.
387 * Return Value: On success, 0 is returned and the block number associated
388 * with @vblocknr is stored in the place pointed by @blocknrp. On error, one
389 * of the following negative error codes is returned.
393 * %-ENOMEM - Insufficient amount of memory available.
395 * %-ENOENT - A block number associated with @vblocknr does not exist.
397 int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
399 struct buffer_head *entry_bh, *bh;
400 struct nilfs_dat_entry *entry;
405 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
409 if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) {
410 bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh);
412 WARN_ON(!buffer_uptodate(bh));
418 kaddr = kmap_atomic(entry_bh->b_page);
419 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
420 blocknr = le64_to_cpu(entry->de_blocknr);
428 kunmap_atomic(kaddr);
433 ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz,
436 struct buffer_head *entry_bh;
437 struct nilfs_dat_entry *entry;
438 struct nilfs_vinfo *vinfo = buf;
441 unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block;
444 for (i = 0; i < nvi; i += n) {
445 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
449 kaddr = kmap_atomic(entry_bh->b_page);
450 /* last virtual block number in this block */
451 first = vinfo->vi_vblocknr;
452 do_div(first, entries_per_block);
453 first *= entries_per_block;
454 last = first + entries_per_block - 1;
456 j < nvi && vinfo->vi_vblocknr >= first &&
457 vinfo->vi_vblocknr <= last;
458 j++, n++, vinfo = (void *)vinfo + visz) {
459 entry = nilfs_palloc_block_get_entry(
460 dat, vinfo->vi_vblocknr, entry_bh, kaddr);
461 vinfo->vi_start = le64_to_cpu(entry->de_start);
462 vinfo->vi_end = le64_to_cpu(entry->de_end);
463 vinfo->vi_blocknr = le64_to_cpu(entry->de_blocknr);
465 kunmap_atomic(kaddr);
473 * nilfs_dat_read - read or get dat inode
474 * @sb: super block instance
475 * @entry_size: size of a dat entry
476 * @raw_inode: on-disk dat inode
477 * @inodep: buffer to store the inode
479 int nilfs_dat_read(struct super_block *sb, size_t entry_size,
480 struct nilfs_inode *raw_inode, struct inode **inodep)
482 static struct lock_class_key dat_lock_key;
484 struct nilfs_dat_info *di;
487 if (entry_size > sb->s_blocksize) {
488 nilfs_err(sb, "too large DAT entry size: %zu bytes",
491 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
492 nilfs_err(sb, "too small DAT entry size: %zu bytes",
497 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
500 if (!(dat->i_state & I_NEW))
503 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
507 err = nilfs_palloc_init_blockgroup(dat, entry_size);
511 di = NILFS_DAT_I(dat);
512 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
513 nilfs_palloc_setup_cache(dat, &di->palloc_cache);
514 err = nilfs_mdt_setup_shadow_map(dat, &di->shadow);
518 err = nilfs_read_inode_common(dat, raw_inode);
522 unlock_new_inode(dat);