Loading...
1/*
2 * dat.c - NILFS disk address translation.
3 *
4 * Copyright (C) 2006-2008 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 *
16 * Written by Koji Sato.
17 */
18
19#include <linux/types.h>
20#include <linux/buffer_head.h>
21#include <linux/string.h>
22#include <linux/errno.h>
23#include "nilfs.h"
24#include "mdt.h"
25#include "alloc.h"
26#include "dat.h"
27
28
29#define NILFS_CNO_MIN ((__u64)1)
30#define NILFS_CNO_MAX (~(__u64)0)
31
32/**
33 * struct nilfs_dat_info - on-memory private data of DAT file
34 * @mi: on-memory private data of metadata file
35 * @palloc_cache: persistent object allocator cache of DAT file
36 * @shadow: shadow map of DAT file
37 */
38struct nilfs_dat_info {
39 struct nilfs_mdt_info mi;
40 struct nilfs_palloc_cache palloc_cache;
41 struct nilfs_shadow_map shadow;
42};
43
44static inline struct nilfs_dat_info *NILFS_DAT_I(struct inode *dat)
45{
46 return (struct nilfs_dat_info *)NILFS_MDT(dat);
47}
48
49static int nilfs_dat_prepare_entry(struct inode *dat,
50 struct nilfs_palloc_req *req, int create)
51{
52 return nilfs_palloc_get_entry_block(dat, req->pr_entry_nr,
53 create, &req->pr_entry_bh);
54}
55
56static void nilfs_dat_commit_entry(struct inode *dat,
57 struct nilfs_palloc_req *req)
58{
59 mark_buffer_dirty(req->pr_entry_bh);
60 nilfs_mdt_mark_dirty(dat);
61 brelse(req->pr_entry_bh);
62}
63
64static void nilfs_dat_abort_entry(struct inode *dat,
65 struct nilfs_palloc_req *req)
66{
67 brelse(req->pr_entry_bh);
68}
69
70int nilfs_dat_prepare_alloc(struct inode *dat, struct nilfs_palloc_req *req)
71{
72 int ret;
73
74 ret = nilfs_palloc_prepare_alloc_entry(dat, req);
75 if (ret < 0)
76 return ret;
77
78 ret = nilfs_dat_prepare_entry(dat, req, 1);
79 if (ret < 0)
80 nilfs_palloc_abort_alloc_entry(dat, req);
81
82 return ret;
83}
84
85void nilfs_dat_commit_alloc(struct inode *dat, struct nilfs_palloc_req *req)
86{
87 struct nilfs_dat_entry *entry;
88 void *kaddr;
89
90 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
91 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
92 req->pr_entry_bh, kaddr);
93 entry->de_start = cpu_to_le64(NILFS_CNO_MIN);
94 entry->de_end = cpu_to_le64(NILFS_CNO_MAX);
95 entry->de_blocknr = cpu_to_le64(0);
96 kunmap_atomic(kaddr);
97
98 nilfs_palloc_commit_alloc_entry(dat, req);
99 nilfs_dat_commit_entry(dat, req);
100}
101
102void nilfs_dat_abort_alloc(struct inode *dat, struct nilfs_palloc_req *req)
103{
104 nilfs_dat_abort_entry(dat, req);
105 nilfs_palloc_abort_alloc_entry(dat, req);
106}
107
108static void nilfs_dat_commit_free(struct inode *dat,
109 struct nilfs_palloc_req *req)
110{
111 struct nilfs_dat_entry *entry;
112 void *kaddr;
113
114 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
115 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
116 req->pr_entry_bh, kaddr);
117 entry->de_start = cpu_to_le64(NILFS_CNO_MIN);
118 entry->de_end = cpu_to_le64(NILFS_CNO_MIN);
119 entry->de_blocknr = cpu_to_le64(0);
120 kunmap_atomic(kaddr);
121
122 nilfs_dat_commit_entry(dat, req);
123 nilfs_palloc_commit_free_entry(dat, req);
124}
125
126int nilfs_dat_prepare_start(struct inode *dat, struct nilfs_palloc_req *req)
127{
128 int ret;
129
130 ret = nilfs_dat_prepare_entry(dat, req, 0);
131 WARN_ON(ret == -ENOENT);
132 return ret;
133}
134
135void nilfs_dat_commit_start(struct inode *dat, struct nilfs_palloc_req *req,
136 sector_t blocknr)
137{
138 struct nilfs_dat_entry *entry;
139 void *kaddr;
140
141 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
142 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
143 req->pr_entry_bh, kaddr);
144 entry->de_start = cpu_to_le64(nilfs_mdt_cno(dat));
145 entry->de_blocknr = cpu_to_le64(blocknr);
146 kunmap_atomic(kaddr);
147
148 nilfs_dat_commit_entry(dat, req);
149}
150
151int nilfs_dat_prepare_end(struct inode *dat, struct nilfs_palloc_req *req)
152{
153 struct nilfs_dat_entry *entry;
154 sector_t blocknr;
155 void *kaddr;
156 int ret;
157
158 ret = nilfs_dat_prepare_entry(dat, req, 0);
159 if (ret < 0) {
160 WARN_ON(ret == -ENOENT);
161 return ret;
162 }
163
164 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
165 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
166 req->pr_entry_bh, kaddr);
167 blocknr = le64_to_cpu(entry->de_blocknr);
168 kunmap_atomic(kaddr);
169
170 if (blocknr == 0) {
171 ret = nilfs_palloc_prepare_free_entry(dat, req);
172 if (ret < 0) {
173 nilfs_dat_abort_entry(dat, req);
174 return ret;
175 }
176 }
177
178 return 0;
179}
180
181void nilfs_dat_commit_end(struct inode *dat, struct nilfs_palloc_req *req,
182 int dead)
183{
184 struct nilfs_dat_entry *entry;
185 __u64 start, end;
186 sector_t blocknr;
187 void *kaddr;
188
189 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
190 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
191 req->pr_entry_bh, kaddr);
192 end = start = le64_to_cpu(entry->de_start);
193 if (!dead) {
194 end = nilfs_mdt_cno(dat);
195 WARN_ON(start > end);
196 }
197 entry->de_end = cpu_to_le64(end);
198 blocknr = le64_to_cpu(entry->de_blocknr);
199 kunmap_atomic(kaddr);
200
201 if (blocknr == 0)
202 nilfs_dat_commit_free(dat, req);
203 else
204 nilfs_dat_commit_entry(dat, req);
205}
206
207void nilfs_dat_abort_end(struct inode *dat, struct nilfs_palloc_req *req)
208{
209 struct nilfs_dat_entry *entry;
210 __u64 start;
211 sector_t blocknr;
212 void *kaddr;
213
214 kaddr = kmap_atomic(req->pr_entry_bh->b_page);
215 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr,
216 req->pr_entry_bh, kaddr);
217 start = le64_to_cpu(entry->de_start);
218 blocknr = le64_to_cpu(entry->de_blocknr);
219 kunmap_atomic(kaddr);
220
221 if (start == nilfs_mdt_cno(dat) && blocknr == 0)
222 nilfs_palloc_abort_free_entry(dat, req);
223 nilfs_dat_abort_entry(dat, req);
224}
225
226int nilfs_dat_prepare_update(struct inode *dat,
227 struct nilfs_palloc_req *oldreq,
228 struct nilfs_palloc_req *newreq)
229{
230 int ret;
231
232 ret = nilfs_dat_prepare_end(dat, oldreq);
233 if (!ret) {
234 ret = nilfs_dat_prepare_alloc(dat, newreq);
235 if (ret < 0)
236 nilfs_dat_abort_end(dat, oldreq);
237 }
238 return ret;
239}
240
241void nilfs_dat_commit_update(struct inode *dat,
242 struct nilfs_palloc_req *oldreq,
243 struct nilfs_palloc_req *newreq, int dead)
244{
245 nilfs_dat_commit_end(dat, oldreq, dead);
246 nilfs_dat_commit_alloc(dat, newreq);
247}
248
249void nilfs_dat_abort_update(struct inode *dat,
250 struct nilfs_palloc_req *oldreq,
251 struct nilfs_palloc_req *newreq)
252{
253 nilfs_dat_abort_end(dat, oldreq);
254 nilfs_dat_abort_alloc(dat, newreq);
255}
256
257/**
258 * nilfs_dat_mark_dirty -
259 * @dat: DAT file inode
260 * @vblocknr: virtual block number
261 *
262 * Description:
263 *
264 * Return Value: On success, 0 is returned. On error, one of the following
265 * negative error codes is returned.
266 *
267 * %-EIO - I/O error.
268 *
269 * %-ENOMEM - Insufficient amount of memory available.
270 */
271int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr)
272{
273 struct nilfs_palloc_req req;
274 int ret;
275
276 req.pr_entry_nr = vblocknr;
277 ret = nilfs_dat_prepare_entry(dat, &req, 0);
278 if (ret == 0)
279 nilfs_dat_commit_entry(dat, &req);
280 return ret;
281}
282
283/**
284 * nilfs_dat_freev - free virtual block numbers
285 * @dat: DAT file inode
286 * @vblocknrs: array of virtual block numbers
287 * @nitems: number of virtual block numbers
288 *
289 * Description: nilfs_dat_freev() frees the virtual block numbers specified by
290 * @vblocknrs and @nitems.
291 *
292 * Return Value: On success, 0 is returned. On error, one of the following
293 * negative error codes is returned.
294 *
295 * %-EIO - I/O error.
296 *
297 * %-ENOMEM - Insufficient amount of memory available.
298 *
299 * %-ENOENT - The virtual block number have not been allocated.
300 */
301int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems)
302{
303 return nilfs_palloc_freev(dat, vblocknrs, nitems);
304}
305
306/**
307 * nilfs_dat_move - change a block number
308 * @dat: DAT file inode
309 * @vblocknr: virtual block number
310 * @blocknr: block number
311 *
312 * Description: nilfs_dat_move() changes the block number associated with
313 * @vblocknr to @blocknr.
314 *
315 * Return Value: On success, 0 is returned. On error, one of the following
316 * negative error codes is returned.
317 *
318 * %-EIO - I/O error.
319 *
320 * %-ENOMEM - Insufficient amount of memory available.
321 */
322int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
323{
324 struct buffer_head *entry_bh;
325 struct nilfs_dat_entry *entry;
326 void *kaddr;
327 int ret;
328
329 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
330 if (ret < 0)
331 return ret;
332
333 /*
334 * The given disk block number (blocknr) is not yet written to
335 * the device at this point.
336 *
337 * To prevent nilfs_dat_translate() from returning the
338 * uncommitted block number, this makes a copy of the entry
339 * buffer and redirects nilfs_dat_translate() to the copy.
340 */
341 if (!buffer_nilfs_redirected(entry_bh)) {
342 ret = nilfs_mdt_freeze_buffer(dat, entry_bh);
343 if (ret) {
344 brelse(entry_bh);
345 return ret;
346 }
347 }
348
349 kaddr = kmap_atomic(entry_bh->b_page);
350 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
351 if (unlikely(entry->de_blocknr == cpu_to_le64(0))) {
352 nilfs_msg(dat->i_sb, KERN_CRIT,
353 "%s: invalid vblocknr = %llu, [%llu, %llu)",
354 __func__, (unsigned long long)vblocknr,
355 (unsigned long long)le64_to_cpu(entry->de_start),
356 (unsigned long long)le64_to_cpu(entry->de_end));
357 kunmap_atomic(kaddr);
358 brelse(entry_bh);
359 return -EINVAL;
360 }
361 WARN_ON(blocknr == 0);
362 entry->de_blocknr = cpu_to_le64(blocknr);
363 kunmap_atomic(kaddr);
364
365 mark_buffer_dirty(entry_bh);
366 nilfs_mdt_mark_dirty(dat);
367
368 brelse(entry_bh);
369
370 return 0;
371}
372
373/**
374 * nilfs_dat_translate - translate a virtual block number to a block number
375 * @dat: DAT file inode
376 * @vblocknr: virtual block number
377 * @blocknrp: pointer to a block number
378 *
379 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr
380 * to the corresponding block number.
381 *
382 * Return Value: On success, 0 is returned and the block number associated
383 * with @vblocknr is stored in the place pointed by @blocknrp. On error, one
384 * of the following negative error codes is returned.
385 *
386 * %-EIO - I/O error.
387 *
388 * %-ENOMEM - Insufficient amount of memory available.
389 *
390 * %-ENOENT - A block number associated with @vblocknr does not exist.
391 */
392int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
393{
394 struct buffer_head *entry_bh, *bh;
395 struct nilfs_dat_entry *entry;
396 sector_t blocknr;
397 void *kaddr;
398 int ret;
399
400 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
401 if (ret < 0)
402 return ret;
403
404 if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) {
405 bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh);
406 if (bh) {
407 WARN_ON(!buffer_uptodate(bh));
408 brelse(entry_bh);
409 entry_bh = bh;
410 }
411 }
412
413 kaddr = kmap_atomic(entry_bh->b_page);
414 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
415 blocknr = le64_to_cpu(entry->de_blocknr);
416 if (blocknr == 0) {
417 ret = -ENOENT;
418 goto out;
419 }
420 *blocknrp = blocknr;
421
422 out:
423 kunmap_atomic(kaddr);
424 brelse(entry_bh);
425 return ret;
426}
427
428ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz,
429 size_t nvi)
430{
431 struct buffer_head *entry_bh;
432 struct nilfs_dat_entry *entry;
433 struct nilfs_vinfo *vinfo = buf;
434 __u64 first, last;
435 void *kaddr;
436 unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block;
437 int i, j, n, ret;
438
439 for (i = 0; i < nvi; i += n) {
440 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
441 0, &entry_bh);
442 if (ret < 0)
443 return ret;
444 kaddr = kmap_atomic(entry_bh->b_page);
445 /* last virtual block number in this block */
446 first = vinfo->vi_vblocknr;
447 do_div(first, entries_per_block);
448 first *= entries_per_block;
449 last = first + entries_per_block - 1;
450 for (j = i, n = 0;
451 j < nvi && vinfo->vi_vblocknr >= first &&
452 vinfo->vi_vblocknr <= last;
453 j++, n++, vinfo = (void *)vinfo + visz) {
454 entry = nilfs_palloc_block_get_entry(
455 dat, vinfo->vi_vblocknr, entry_bh, kaddr);
456 vinfo->vi_start = le64_to_cpu(entry->de_start);
457 vinfo->vi_end = le64_to_cpu(entry->de_end);
458 vinfo->vi_blocknr = le64_to_cpu(entry->de_blocknr);
459 }
460 kunmap_atomic(kaddr);
461 brelse(entry_bh);
462 }
463
464 return nvi;
465}
466
467/**
468 * nilfs_dat_read - read or get dat inode
469 * @sb: super block instance
470 * @entry_size: size of a dat entry
471 * @raw_inode: on-disk dat inode
472 * @inodep: buffer to store the inode
473 */
474int nilfs_dat_read(struct super_block *sb, size_t entry_size,
475 struct nilfs_inode *raw_inode, struct inode **inodep)
476{
477 static struct lock_class_key dat_lock_key;
478 struct inode *dat;
479 struct nilfs_dat_info *di;
480 int err;
481
482 if (entry_size > sb->s_blocksize) {
483 nilfs_msg(sb, KERN_ERR, "too large DAT entry size: %zu bytes",
484 entry_size);
485 return -EINVAL;
486 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
487 nilfs_msg(sb, KERN_ERR, "too small DAT entry size: %zu bytes",
488 entry_size);
489 return -EINVAL;
490 }
491
492 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
493 if (unlikely(!dat))
494 return -ENOMEM;
495 if (!(dat->i_state & I_NEW))
496 goto out;
497
498 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
499 if (err)
500 goto failed;
501
502 err = nilfs_palloc_init_blockgroup(dat, entry_size);
503 if (err)
504 goto failed;
505
506 di = NILFS_DAT_I(dat);
507 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
508 nilfs_palloc_setup_cache(dat, &di->palloc_cache);
509 nilfs_mdt_setup_shadow_map(dat, &di->shadow);
510
511 err = nilfs_read_inode_common(dat, raw_inode);
512 if (err)
513 goto failed;
514
515 unlock_new_inode(dat);
516 out:
517 *inodep = dat;
518 return 0;
519 failed:
520 iget_failed(dat);
521 return err;
522}
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * NILFS disk address translation.
4 *
5 * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Written by Koji Sato.
8 */
9
10#include <linux/types.h>
11#include <linux/buffer_head.h>
12#include <linux/string.h>
13#include <linux/errno.h>
14#include "nilfs.h"
15#include "mdt.h"
16#include "alloc.h"
17#include "dat.h"
18
19
20#define NILFS_CNO_MIN ((__u64)1)
21#define NILFS_CNO_MAX (~(__u64)0)
22
23/**
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
28 */
29struct nilfs_dat_info {
30 struct nilfs_mdt_info mi;
31 struct nilfs_palloc_cache palloc_cache;
32 struct nilfs_shadow_map shadow;
33};
34
35static inline struct nilfs_dat_info *NILFS_DAT_I(struct inode *dat)
36{
37 return (struct nilfs_dat_info *)NILFS_MDT(dat);
38}
39
40static int nilfs_dat_prepare_entry(struct inode *dat,
41 struct nilfs_palloc_req *req, int create)
42{
43 int ret;
44
45 ret = nilfs_palloc_get_entry_block(dat, req->pr_entry_nr,
46 create, &req->pr_entry_bh);
47 if (unlikely(ret == -ENOENT)) {
48 nilfs_err(dat->i_sb,
49 "DAT doesn't have a block to manage vblocknr = %llu",
50 (unsigned long long)req->pr_entry_nr);
51 /*
52 * Return internal code -EINVAL to notify bmap layer of
53 * metadata corruption.
54 */
55 ret = -EINVAL;
56 }
57 return ret;
58}
59
60static void nilfs_dat_commit_entry(struct inode *dat,
61 struct nilfs_palloc_req *req)
62{
63 mark_buffer_dirty(req->pr_entry_bh);
64 nilfs_mdt_mark_dirty(dat);
65 brelse(req->pr_entry_bh);
66}
67
68static void nilfs_dat_abort_entry(struct inode *dat,
69 struct nilfs_palloc_req *req)
70{
71 brelse(req->pr_entry_bh);
72}
73
74int nilfs_dat_prepare_alloc(struct inode *dat, struct nilfs_palloc_req *req)
75{
76 int ret;
77
78 ret = nilfs_palloc_prepare_alloc_entry(dat, req, true);
79 if (ret < 0)
80 return ret;
81
82 ret = nilfs_dat_prepare_entry(dat, req, 1);
83 if (ret < 0)
84 nilfs_palloc_abort_alloc_entry(dat, req);
85
86 return ret;
87}
88
89void nilfs_dat_commit_alloc(struct inode *dat, struct nilfs_palloc_req *req)
90{
91 struct nilfs_dat_entry *entry;
92 size_t offset;
93
94 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
95 req->pr_entry_bh);
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);
100 kunmap_local(entry);
101
102 nilfs_palloc_commit_alloc_entry(dat, req);
103 nilfs_dat_commit_entry(dat, req);
104}
105
106void nilfs_dat_abort_alloc(struct inode *dat, struct nilfs_palloc_req *req)
107{
108 nilfs_dat_abort_entry(dat, req);
109 nilfs_palloc_abort_alloc_entry(dat, req);
110}
111
112static void nilfs_dat_commit_free(struct inode *dat,
113 struct nilfs_palloc_req *req)
114{
115 struct nilfs_dat_entry *entry;
116 size_t offset;
117
118 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
119 req->pr_entry_bh);
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);
124 kunmap_local(entry);
125
126 nilfs_dat_commit_entry(dat, req);
127
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);
132 return;
133 }
134 nilfs_palloc_commit_free_entry(dat, req);
135}
136
137int nilfs_dat_prepare_start(struct inode *dat, struct nilfs_palloc_req *req)
138{
139 return nilfs_dat_prepare_entry(dat, req, 0);
140}
141
142void nilfs_dat_commit_start(struct inode *dat, struct nilfs_palloc_req *req,
143 sector_t blocknr)
144{
145 struct nilfs_dat_entry *entry;
146 size_t offset;
147
148 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
149 req->pr_entry_bh);
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);
153 kunmap_local(entry);
154
155 nilfs_dat_commit_entry(dat, req);
156}
157
158int nilfs_dat_prepare_end(struct inode *dat, struct nilfs_palloc_req *req)
159{
160 struct nilfs_dat_entry *entry;
161 __u64 start;
162 sector_t blocknr;
163 size_t offset;
164 int ret;
165
166 ret = nilfs_dat_prepare_entry(dat, req, 0);
167 if (ret < 0)
168 return ret;
169
170 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
171 req->pr_entry_bh);
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);
175 kunmap_local(entry);
176
177 if (blocknr == 0) {
178 ret = nilfs_palloc_prepare_free_entry(dat, req);
179 if (ret < 0) {
180 nilfs_dat_abort_entry(dat, req);
181 return ret;
182 }
183 }
184 if (unlikely(start > nilfs_mdt_cno(dat))) {
185 nilfs_err(dat->i_sb,
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);
191 return -EINVAL;
192 }
193
194 return 0;
195}
196
197void nilfs_dat_commit_end(struct inode *dat, struct nilfs_palloc_req *req,
198 int dead)
199{
200 struct nilfs_dat_entry *entry;
201 __u64 start, end;
202 sector_t blocknr;
203 size_t offset;
204
205 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
206 req->pr_entry_bh);
207 entry = kmap_local_folio(req->pr_entry_bh->b_folio, offset);
208 end = start = le64_to_cpu(entry->de_start);
209 if (!dead) {
210 end = nilfs_mdt_cno(dat);
211 WARN_ON(start > end);
212 }
213 entry->de_end = cpu_to_le64(end);
214 blocknr = le64_to_cpu(entry->de_blocknr);
215 kunmap_local(entry);
216
217 if (blocknr == 0)
218 nilfs_dat_commit_free(dat, req);
219 else
220 nilfs_dat_commit_entry(dat, req);
221}
222
223void nilfs_dat_abort_end(struct inode *dat, struct nilfs_palloc_req *req)
224{
225 struct nilfs_dat_entry *entry;
226 __u64 start;
227 sector_t blocknr;
228 size_t offset;
229
230 offset = nilfs_palloc_entry_offset(dat, req->pr_entry_nr,
231 req->pr_entry_bh);
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);
235 kunmap_local(entry);
236
237 if (start == nilfs_mdt_cno(dat) && blocknr == 0)
238 nilfs_palloc_abort_free_entry(dat, req);
239 nilfs_dat_abort_entry(dat, req);
240}
241
242int nilfs_dat_prepare_update(struct inode *dat,
243 struct nilfs_palloc_req *oldreq,
244 struct nilfs_palloc_req *newreq)
245{
246 int ret;
247
248 ret = nilfs_dat_prepare_end(dat, oldreq);
249 if (!ret) {
250 ret = nilfs_dat_prepare_alloc(dat, newreq);
251 if (ret < 0)
252 nilfs_dat_abort_end(dat, oldreq);
253 }
254 return ret;
255}
256
257void nilfs_dat_commit_update(struct inode *dat,
258 struct nilfs_palloc_req *oldreq,
259 struct nilfs_palloc_req *newreq, int dead)
260{
261 nilfs_dat_commit_end(dat, oldreq, dead);
262 nilfs_dat_commit_alloc(dat, newreq);
263}
264
265void nilfs_dat_abort_update(struct inode *dat,
266 struct nilfs_palloc_req *oldreq,
267 struct nilfs_palloc_req *newreq)
268{
269 nilfs_dat_abort_end(dat, oldreq);
270 nilfs_dat_abort_alloc(dat, newreq);
271}
272
273/**
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
278 *
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.
283 */
284int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr)
285{
286 struct nilfs_palloc_req req;
287 int ret;
288
289 req.pr_entry_nr = vblocknr;
290 ret = nilfs_dat_prepare_entry(dat, &req, 0);
291 if (ret == 0)
292 nilfs_dat_commit_entry(dat, &req);
293 return ret;
294}
295
296/**
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
301 *
302 * Description: nilfs_dat_freev() frees the virtual block numbers specified by
303 * @vblocknrs and @nitems.
304 *
305 * Return Value: On success, 0 is returned. On error, one of the following
306 * negative error codes is returned.
307 *
308 * %-EIO - I/O error.
309 *
310 * %-ENOMEM - Insufficient amount of memory available.
311 *
312 * %-ENOENT - The virtual block number have not been allocated.
313 */
314int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems)
315{
316 return nilfs_palloc_freev(dat, vblocknrs, nitems);
317}
318
319/**
320 * nilfs_dat_move - change a block number
321 * @dat: DAT file inode
322 * @vblocknr: virtual block number
323 * @blocknr: block number
324 *
325 * Description: nilfs_dat_move() changes the block number associated with
326 * @vblocknr to @blocknr.
327 *
328 * Return Value: On success, 0 is returned. On error, one of the following
329 * negative error codes is returned.
330 *
331 * %-EIO - I/O error.
332 *
333 * %-ENOMEM - Insufficient amount of memory available.
334 */
335int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
336{
337 struct buffer_head *entry_bh;
338 struct nilfs_dat_entry *entry;
339 size_t offset;
340 int ret;
341
342 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
343 if (ret < 0)
344 return ret;
345
346 /*
347 * The given disk block number (blocknr) is not yet written to
348 * the device at this point.
349 *
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.
353 */
354 if (!buffer_nilfs_redirected(entry_bh)) {
355 ret = nilfs_mdt_freeze_buffer(dat, entry_bh);
356 if (ret) {
357 brelse(entry_bh);
358 return ret;
359 }
360 }
361
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));
370 kunmap_local(entry);
371 brelse(entry_bh);
372 return -EINVAL;
373 }
374 WARN_ON(blocknr == 0);
375 entry->de_blocknr = cpu_to_le64(blocknr);
376 kunmap_local(entry);
377
378 mark_buffer_dirty(entry_bh);
379 nilfs_mdt_mark_dirty(dat);
380
381 brelse(entry_bh);
382
383 return 0;
384}
385
386/**
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
391 *
392 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr
393 * to the corresponding block number.
394 *
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.
398 *
399 * %-EIO - I/O error.
400 *
401 * %-ENOMEM - Insufficient amount of memory available.
402 *
403 * %-ENOENT - A block number associated with @vblocknr does not exist.
404 */
405int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
406{
407 struct buffer_head *entry_bh, *bh;
408 struct nilfs_dat_entry *entry;
409 sector_t blocknr;
410 size_t offset;
411 int ret;
412
413 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
414 if (ret < 0)
415 return ret;
416
417 if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) {
418 bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh);
419 if (bh) {
420 WARN_ON(!buffer_uptodate(bh));
421 brelse(entry_bh);
422 entry_bh = bh;
423 }
424 }
425
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);
429 if (blocknr == 0) {
430 ret = -ENOENT;
431 goto out;
432 }
433 *blocknrp = blocknr;
434
435 out:
436 kunmap_local(entry);
437 brelse(entry_bh);
438 return ret;
439}
440
441ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz,
442 size_t nvi)
443{
444 struct buffer_head *entry_bh;
445 struct nilfs_dat_entry *entry, *first_entry;
446 struct nilfs_vinfo *vinfo = buf;
447 __u64 first, last;
448 size_t offset;
449 unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block;
450 unsigned int entry_size = NILFS_MDT(dat)->mi_entry_size;
451 int i, j, n, ret;
452
453 for (i = 0; i < nvi; i += n) {
454 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
455 0, &entry_bh);
456 if (ret < 0)
457 return ret;
458
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 */
463
464 last = first + entries_per_block - 1;
465 /* last virtual block number in this block */
466
467 offset = nilfs_palloc_entry_offset(dat, first, entry_bh);
468 first_entry = kmap_local_folio(entry_bh->b_folio, offset);
469 for (j = i, n = 0;
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);
478 }
479 kunmap_local(first_entry);
480 brelse(entry_bh);
481 }
482
483 return nvi;
484}
485
486/**
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
492 */
493int nilfs_dat_read(struct super_block *sb, size_t entry_size,
494 struct nilfs_inode *raw_inode, struct inode **inodep)
495{
496 static struct lock_class_key dat_lock_key;
497 struct inode *dat;
498 struct nilfs_dat_info *di;
499 int err;
500
501 if (entry_size > sb->s_blocksize) {
502 nilfs_err(sb, "too large DAT entry size: %zu bytes",
503 entry_size);
504 return -EINVAL;
505 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
506 nilfs_err(sb, "too small DAT entry size: %zu bytes",
507 entry_size);
508 return -EINVAL;
509 }
510
511 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
512 if (unlikely(!dat))
513 return -ENOMEM;
514 if (!(dat->i_state & I_NEW))
515 goto out;
516
517 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
518 if (err)
519 goto failed;
520
521 err = nilfs_palloc_init_blockgroup(dat, entry_size);
522 if (err)
523 goto failed;
524
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);
529 if (err)
530 goto failed;
531
532 err = nilfs_read_inode_common(dat, raw_inode);
533 if (err)
534 goto failed;
535
536 unlock_new_inode(dat);
537 out:
538 *inodep = dat;
539 return 0;
540 failed:
541 iget_failed(dat);
542 return err;
543}