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 */
29 struct nilfs_dat_info {
30 struct nilfs_mdt_info mi;
31 struct nilfs_palloc_cache palloc_cache;
32 struct nilfs_shadow_map shadow;
33 };
34
NILFS_DAT_I(struct inode * dat)35 static inline struct nilfs_dat_info *NILFS_DAT_I(struct inode *dat)
36 {
37 return (struct nilfs_dat_info *)NILFS_MDT(dat);
38 }
39
nilfs_dat_prepare_entry(struct inode * dat,struct nilfs_palloc_req * req,int create)40 static 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
nilfs_dat_commit_entry(struct inode * dat,struct nilfs_palloc_req * req)60 static 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
nilfs_dat_abort_entry(struct inode * dat,struct nilfs_palloc_req * req)68 static void nilfs_dat_abort_entry(struct inode *dat,
69 struct nilfs_palloc_req *req)
70 {
71 brelse(req->pr_entry_bh);
72 }
73
nilfs_dat_prepare_alloc(struct inode * dat,struct nilfs_palloc_req * req)74 int 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
nilfs_dat_commit_alloc(struct inode * dat,struct nilfs_palloc_req * req)89 void 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_local_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_local(kaddr);
101
102 nilfs_palloc_commit_alloc_entry(dat, req);
103 nilfs_dat_commit_entry(dat, req);
104 }
105
nilfs_dat_abort_alloc(struct inode * dat,struct nilfs_palloc_req * req)106 void 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
nilfs_dat_commit_free(struct inode * dat,struct nilfs_palloc_req * req)112 static 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_local_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_local(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
nilfs_dat_prepare_start(struct inode * dat,struct nilfs_palloc_req * req)137 int nilfs_dat_prepare_start(struct inode *dat, struct nilfs_palloc_req *req)
138 {
139 return nilfs_dat_prepare_entry(dat, req, 0);
140 }
141
nilfs_dat_commit_start(struct inode * dat,struct nilfs_palloc_req * req,sector_t blocknr)142 void 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_local_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_local(kaddr);
154
155 nilfs_dat_commit_entry(dat, req);
156 }
157
nilfs_dat_prepare_end(struct inode * dat,struct nilfs_palloc_req * req)158 int 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, 0);
167 if (ret < 0)
168 return ret;
169
170 kaddr = kmap_local_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_local(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
nilfs_dat_commit_end(struct inode * dat,struct nilfs_palloc_req * req,int dead)197 void 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_local_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(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(kaddr);
216
217 if (blocknr == 0)
218 nilfs_dat_commit_free(dat, req);
219 else
220 nilfs_dat_commit_entry(dat, req);
221 }
222
nilfs_dat_abort_end(struct inode * dat,struct nilfs_palloc_req * req)223 void 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_local_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_local(kaddr);
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
nilfs_dat_prepare_update(struct inode * dat,struct nilfs_palloc_req * oldreq,struct nilfs_palloc_req * newreq)242 int 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
nilfs_dat_commit_update(struct inode * dat,struct nilfs_palloc_req * oldreq,struct nilfs_palloc_req * newreq,int dead)257 void 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
nilfs_dat_abort_update(struct inode * dat,struct nilfs_palloc_req * oldreq,struct nilfs_palloc_req * newreq)265 void 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 */
nilfs_dat_mark_dirty(struct inode * dat,__u64 vblocknr)284 int 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 */
nilfs_dat_freev(struct inode * dat,__u64 * vblocknrs,size_t nitems)314 int 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 */
nilfs_dat_move(struct inode * dat,__u64 vblocknr,sector_t blocknr)335 int 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 void *kaddr;
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 kaddr = kmap_local_page(entry_bh->b_page);
363 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
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(kaddr);
371 brelse(entry_bh);
372 return -EINVAL;
373 }
374 WARN_ON(blocknr == 0);
375 entry->de_blocknr = cpu_to_le64(blocknr);
376 kunmap_local(kaddr);
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 */
nilfs_dat_translate(struct inode * dat,__u64 vblocknr,sector_t * blocknrp)405 int 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 void *kaddr;
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 kaddr = kmap_local_page(entry_bh->b_page);
427 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
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(kaddr);
437 brelse(entry_bh);
438 return ret;
439 }
440
nilfs_dat_get_vinfo(struct inode * dat,void * buf,unsigned int visz,size_t nvi)441 ssize_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;
446 struct nilfs_vinfo *vinfo = buf;
447 __u64 first, last;
448 void *kaddr;
449 unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block;
450 int i, j, n, ret;
451
452 for (i = 0; i < nvi; i += n) {
453 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
454 0, &entry_bh);
455 if (ret < 0)
456 return ret;
457 kaddr = kmap_local_page(entry_bh->b_page);
458 /* last virtual block number in this block */
459 first = vinfo->vi_vblocknr;
460 first = div64_ul(first, entries_per_block);
461 first *= entries_per_block;
462 last = first + entries_per_block - 1;
463 for (j = i, n = 0;
464 j < nvi && vinfo->vi_vblocknr >= first &&
465 vinfo->vi_vblocknr <= last;
466 j++, n++, vinfo = (void *)vinfo + visz) {
467 entry = nilfs_palloc_block_get_entry(
468 dat, vinfo->vi_vblocknr, entry_bh, kaddr);
469 vinfo->vi_start = le64_to_cpu(entry->de_start);
470 vinfo->vi_end = le64_to_cpu(entry->de_end);
471 vinfo->vi_blocknr = le64_to_cpu(entry->de_blocknr);
472 }
473 kunmap_local(kaddr);
474 brelse(entry_bh);
475 }
476
477 return nvi;
478 }
479
480 /**
481 * nilfs_dat_read - read or get dat inode
482 * @sb: super block instance
483 * @entry_size: size of a dat entry
484 * @raw_inode: on-disk dat inode
485 * @inodep: buffer to store the inode
486 */
nilfs_dat_read(struct super_block * sb,size_t entry_size,struct nilfs_inode * raw_inode,struct inode ** inodep)487 int nilfs_dat_read(struct super_block *sb, size_t entry_size,
488 struct nilfs_inode *raw_inode, struct inode **inodep)
489 {
490 static struct lock_class_key dat_lock_key;
491 struct inode *dat;
492 struct nilfs_dat_info *di;
493 int err;
494
495 if (entry_size > sb->s_blocksize) {
496 nilfs_err(sb, "too large DAT entry size: %zu bytes",
497 entry_size);
498 return -EINVAL;
499 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
500 nilfs_err(sb, "too small DAT entry size: %zu bytes",
501 entry_size);
502 return -EINVAL;
503 }
504
505 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
506 if (unlikely(!dat))
507 return -ENOMEM;
508 if (!(dat->i_state & I_NEW))
509 goto out;
510
511 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
512 if (err)
513 goto failed;
514
515 err = nilfs_palloc_init_blockgroup(dat, entry_size);
516 if (err)
517 goto failed;
518
519 di = NILFS_DAT_I(dat);
520 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
521 nilfs_palloc_setup_cache(dat, &di->palloc_cache);
522 err = nilfs_mdt_setup_shadow_map(dat, &di->shadow);
523 if (err)
524 goto failed;
525
526 err = nilfs_read_inode_common(dat, raw_inode);
527 if (err)
528 goto failed;
529
530 unlock_new_inode(dat);
531 out:
532 *inodep = dat;
533 return 0;
534 failed:
535 iget_failed(dat);
536 return err;
537 }
538