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(inode: 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(bh: req->pr_entry_bh);
64 nilfs_mdt_mark_dirty(inode: dat);
65 brelse(bh: req->pr_entry_bh);
66}
67
68static void nilfs_dat_abort_entry(struct inode *dat,
69 struct nilfs_palloc_req *req)
70{
71 brelse(bh: 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);
79 if (ret < 0)
80 return ret;
81
82 ret = nilfs_dat_prepare_entry(dat, req, create: 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 void *kaddr;
93
94 kaddr = kmap_atomic(page: 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);
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 void *kaddr;
117
118 kaddr = kmap_atomic(page: 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);
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, create: 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 void *kaddr;
147
148 kaddr = kmap_atomic(page: 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);
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 void *kaddr;
164 int ret;
165
166 ret = nilfs_dat_prepare_entry(dat, req, create: 0);
167 if (ret < 0)
168 return ret;
169
170 kaddr = kmap_atomic(page: 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);
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 void *kaddr;
204
205 kaddr = kmap_atomic(page: 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);
209 if (!dead) {
210 end = nilfs_mdt_cno(inode: 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_atomic(kaddr);
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 void *kaddr;
229
230 kaddr = kmap_atomic(page: 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);
236
237 if (start == nilfs_mdt_cno(inode: 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, req: oldreq);
249 if (!ret) {
250 ret = nilfs_dat_prepare_alloc(dat, req: newreq);
251 if (ret < 0)
252 nilfs_dat_abort_end(dat, req: 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, req: oldreq, dead);
262 nilfs_dat_commit_alloc(dat, req: 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, req: oldreq);
270 nilfs_dat_abort_alloc(dat, req: newreq);
271}
272
273/**
274 * nilfs_dat_mark_dirty -
275 * @dat: DAT file inode
276 * @vblocknr: virtual block number
277 *
278 * Description:
279 *
280 * Return Value: On success, 0 is returned. On error, one of the following
281 * negative error codes is returned.
282 *
283 * %-EIO - I/O error.
284 *
285 * %-ENOMEM - Insufficient amount of memory available.
286 */
287int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr)
288{
289 struct nilfs_palloc_req req;
290 int ret;
291
292 req.pr_entry_nr = vblocknr;
293 ret = nilfs_dat_prepare_entry(dat, req: &req, create: 0);
294 if (ret == 0)
295 nilfs_dat_commit_entry(dat, req: &req);
296 return ret;
297}
298
299/**
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
304 *
305 * Description: nilfs_dat_freev() frees the virtual block numbers specified by
306 * @vblocknrs and @nitems.
307 *
308 * Return Value: On success, 0 is returned. On error, one of the following
309 * negative error codes is returned.
310 *
311 * %-EIO - I/O error.
312 *
313 * %-ENOMEM - Insufficient amount of memory available.
314 *
315 * %-ENOENT - The virtual block number have not been allocated.
316 */
317int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems)
318{
319 return nilfs_palloc_freev(dat, vblocknrs, nitems);
320}
321
322/**
323 * nilfs_dat_move - change a block number
324 * @dat: DAT file inode
325 * @vblocknr: virtual block number
326 * @blocknr: block number
327 *
328 * Description: nilfs_dat_move() changes the block number associated with
329 * @vblocknr to @blocknr.
330 *
331 * Return Value: On success, 0 is returned. On error, one of the following
332 * negative error codes is returned.
333 *
334 * %-EIO - I/O error.
335 *
336 * %-ENOMEM - Insufficient amount of memory available.
337 */
338int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
339{
340 struct buffer_head *entry_bh;
341 struct nilfs_dat_entry *entry;
342 void *kaddr;
343 int ret;
344
345 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
346 if (ret < 0)
347 return ret;
348
349 /*
350 * The given disk block number (blocknr) is not yet written to
351 * the device at this point.
352 *
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.
356 */
357 if (!buffer_nilfs_redirected(bh: entry_bh)) {
358 ret = nilfs_mdt_freeze_buffer(inode: dat, bh: entry_bh);
359 if (ret) {
360 brelse(bh: entry_bh);
361 return ret;
362 }
363 }
364
365 kaddr = kmap_atomic(page: 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);
374 brelse(bh: entry_bh);
375 return -EINVAL;
376 }
377 WARN_ON(blocknr == 0);
378 entry->de_blocknr = cpu_to_le64(blocknr);
379 kunmap_atomic(kaddr);
380
381 mark_buffer_dirty(bh: entry_bh);
382 nilfs_mdt_mark_dirty(inode: dat);
383
384 brelse(bh: entry_bh);
385
386 return 0;
387}
388
389/**
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
394 *
395 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr
396 * to the corresponding block number.
397 *
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.
401 *
402 * %-EIO - I/O error.
403 *
404 * %-ENOMEM - Insufficient amount of memory available.
405 *
406 * %-ENOENT - A block number associated with @vblocknr does not exist.
407 */
408int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
409{
410 struct buffer_head *entry_bh, *bh;
411 struct nilfs_dat_entry *entry;
412 sector_t blocknr;
413 void *kaddr;
414 int ret;
415
416 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
417 if (ret < 0)
418 return ret;
419
420 if (!nilfs_doing_gc() && buffer_nilfs_redirected(bh: entry_bh)) {
421 bh = nilfs_mdt_get_frozen_buffer(inode: dat, bh: entry_bh);
422 if (bh) {
423 WARN_ON(!buffer_uptodate(bh));
424 brelse(bh: entry_bh);
425 entry_bh = bh;
426 }
427 }
428
429 kaddr = kmap_atomic(page: entry_bh->b_page);
430 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
431 blocknr = le64_to_cpu(entry->de_blocknr);
432 if (blocknr == 0) {
433 ret = -ENOENT;
434 goto out;
435 }
436 *blocknrp = blocknr;
437
438 out:
439 kunmap_atomic(kaddr);
440 brelse(bh: entry_bh);
441 return ret;
442}
443
444ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz,
445 size_t nvi)
446{
447 struct buffer_head *entry_bh;
448 struct nilfs_dat_entry *entry;
449 struct nilfs_vinfo *vinfo = buf;
450 __u64 first, last;
451 void *kaddr;
452 unsigned long entries_per_block = NILFS_MDT(inode: dat)->mi_entries_per_block;
453 int i, j, n, ret;
454
455 for (i = 0; i < nvi; i += n) {
456 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
457 0, &entry_bh);
458 if (ret < 0)
459 return ret;
460 kaddr = kmap_atomic(page: 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;
466 for (j = i, n = 0;
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);
475 }
476 kunmap_atomic(kaddr);
477 brelse(bh: entry_bh);
478 }
479
480 return nvi;
481}
482
483/**
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
489 */
490int nilfs_dat_read(struct super_block *sb, size_t entry_size,
491 struct nilfs_inode *raw_inode, struct inode **inodep)
492{
493 static struct lock_class_key dat_lock_key;
494 struct inode *dat;
495 struct nilfs_dat_info *di;
496 int err;
497
498 if (entry_size > sb->s_blocksize) {
499 nilfs_err(sb, "too large DAT entry size: %zu bytes",
500 entry_size);
501 return -EINVAL;
502 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
503 nilfs_err(sb, "too small DAT entry size: %zu bytes",
504 entry_size);
505 return -EINVAL;
506 }
507
508 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
509 if (unlikely(!dat))
510 return -ENOMEM;
511 if (!(dat->i_state & I_NEW))
512 goto out;
513
514 err = nilfs_mdt_init(inode: dat, NILFS_MDT_GFP, objsz: sizeof(*di));
515 if (err)
516 goto failed;
517
518 err = nilfs_palloc_init_blockgroup(dat, entry_size);
519 if (err)
520 goto failed;
521
522 di = NILFS_DAT_I(dat);
523 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
524 nilfs_palloc_setup_cache(inode: dat, cache: &di->palloc_cache);
525 err = nilfs_mdt_setup_shadow_map(inode: dat, shadow: &di->shadow);
526 if (err)
527 goto failed;
528
529 err = nilfs_read_inode_common(dat, raw_inode);
530 if (err)
531 goto failed;
532
533 unlock_new_inode(dat);
534 out:
535 *inodep = dat;
536 return 0;
537 failed:
538 iget_failed(dat);
539 return err;
540}
541

source code of linux/fs/nilfs2/dat.c