dat.c

  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 one of the following negative error codes on
280 * failure:
281 * * %-EINVAL	- Invalid DAT entry (internal code).
282 * * %-EIO	- I/O error (including metadata corruption).
283 * * %-ENOMEM	- Insufficient memory available.
284 */
285int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr)
286{
287	struct nilfs_palloc_req req;
288	int ret;
289
290	req.pr_entry_nr = vblocknr;
291	ret = nilfs_dat_prepare_entry(dat, &req, 0);
292	if (ret == 0)
293		nilfs_dat_commit_entry(dat, &req);
294	return ret;
295}
296
297/**
298 * nilfs_dat_freev - free virtual block numbers
299 * @dat: DAT file inode
300 * @vblocknrs: array of virtual block numbers
301 * @nitems: number of virtual block numbers
302 *
303 * Description: nilfs_dat_freev() frees the virtual block numbers specified by
304 * @vblocknrs and @nitems.
305 *
306 * Return: 0 on success, or one of the following negative error codes on
307 * failure:
308 * * %-EIO	- I/O error (including metadata corruption).
309 * * %-ENOENT	- The virtual block number have not been allocated.
310 * * %-ENOMEM	- Insufficient memory available.
311 */
312int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems)
313{
314	return nilfs_palloc_freev(dat, vblocknrs, nitems);
315}
316
317/**
318 * nilfs_dat_move - change a block number
319 * @dat: DAT file inode
320 * @vblocknr: virtual block number
321 * @blocknr: block number
322 *
323 * Description: nilfs_dat_move() changes the block number associated with
324 * @vblocknr to @blocknr.
325 *
326 * Return: 0 on success, or one of the following negative error codes on
327 * failure:
328 * * %-EIO	- I/O error (including metadata corruption).
329 * * %-ENOMEM	- Insufficient memory available.
330 */
331int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
332{
333	struct buffer_head *entry_bh;
334	struct nilfs_dat_entry *entry;
335	size_t offset;
336	int ret;
337
338	ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
339	if (ret < 0)
340		return ret;
341
342	/*
343	 * The given disk block number (blocknr) is not yet written to
344	 * the device at this point.
345	 *
346	 * To prevent nilfs_dat_translate() from returning the
347	 * uncommitted block number, this makes a copy of the entry
348	 * buffer and redirects nilfs_dat_translate() to the copy.
349	 */
350	if (!buffer_nilfs_redirected(entry_bh)) {
351		ret = nilfs_mdt_freeze_buffer(dat, entry_bh);
352		if (ret) {
353			brelse(entry_bh);
354			return ret;
355		}
356	}
357
358	offset = nilfs_palloc_entry_offset(dat, vblocknr, entry_bh);
359	entry = kmap_local_folio(entry_bh->b_folio, offset);
360	if (unlikely(entry->de_blocknr == cpu_to_le64(0))) {
361		nilfs_crit(dat->i_sb,
362			   "%s: invalid vblocknr = %llu, [%llu, %llu)",
363			   __func__, (unsigned long long)vblocknr,
364			   (unsigned long long)le64_to_cpu(entry->de_start),
365			   (unsigned long long)le64_to_cpu(entry->de_end));
366		kunmap_local(entry);
367		brelse(entry_bh);
368		return -EINVAL;
369	}
370	WARN_ON(blocknr == 0);
371	entry->de_blocknr = cpu_to_le64(blocknr);
372	kunmap_local(entry);
373
374	mark_buffer_dirty(entry_bh);
375	nilfs_mdt_mark_dirty(dat);
376
377	brelse(entry_bh);
378
379	return 0;
380}
381
382/**
383 * nilfs_dat_translate - translate a virtual block number to a block number
384 * @dat: DAT file inode
385 * @vblocknr: virtual block number
386 * @blocknrp: pointer to a block number
387 *
388 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr
389 * to the corresponding block number.  The block number associated with
390 * @vblocknr is stored in the place pointed to by @blocknrp.
391 *
392 * Return: 0 on success, or one of the following negative error codes on
393 * failure:
394 * * %-EIO	- I/O error (including metadata corruption).
395 * * %-ENOENT	- A block number associated with @vblocknr does not exist.
396 * * %-ENOMEM	- Insufficient memory available.
397 */
398int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
399{
400	struct buffer_head *entry_bh, *bh;
401	struct nilfs_dat_entry *entry;
402	sector_t blocknr;
403	size_t offset;
404	int ret;
405
406	ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
407	if (ret < 0)
408		return ret;
409
410	if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) {
411		bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh);
412		if (bh) {
413			WARN_ON(!buffer_uptodate(bh));
414			brelse(entry_bh);
415			entry_bh = bh;
416		}
417	}
418
419	offset = nilfs_palloc_entry_offset(dat, vblocknr, entry_bh);
420	entry = kmap_local_folio(entry_bh->b_folio, offset);
421	blocknr = le64_to_cpu(entry->de_blocknr);
422	if (blocknr == 0) {
423		ret = -ENOENT;
424		goto out;
425	}
426	*blocknrp = blocknr;
427
428 out:
429	kunmap_local(entry);
430	brelse(entry_bh);
431	return ret;
432}
433
434ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz,
435			    size_t nvi)
436{
437	struct buffer_head *entry_bh;
438	struct nilfs_dat_entry *entry, *first_entry;
439	struct nilfs_vinfo *vinfo = buf;
440	__u64 first, last;
441	size_t offset;
442	unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block;
443	unsigned int entry_size = NILFS_MDT(dat)->mi_entry_size;
444	int i, j, n, ret;
445
446	for (i = 0; i < nvi; i += n) {
447		ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr,
448						   0, &entry_bh);
449		if (ret < 0)
450			return ret;
451
452		first = vinfo->vi_vblocknr;
453		first = div64_ul(first, entries_per_block);
454		first *= entries_per_block;
455		/* first virtual block number in this block */
456
457		last = first + entries_per_block - 1;
458		/* last virtual block number in this block */
459
460		offset = nilfs_palloc_entry_offset(dat, first, entry_bh);
461		first_entry = kmap_local_folio(entry_bh->b_folio, offset);
462		for (j = i, n = 0;
463		     j < nvi && vinfo->vi_vblocknr >= first &&
464			     vinfo->vi_vblocknr <= last;
465		     j++, n++, vinfo = (void *)vinfo + visz) {
466			entry = (void *)first_entry +
467				(vinfo->vi_vblocknr - first) * entry_size;
468			vinfo->vi_start = le64_to_cpu(entry->de_start);
469			vinfo->vi_end = le64_to_cpu(entry->de_end);
470			vinfo->vi_blocknr = le64_to_cpu(entry->de_blocknr);
471		}
472		kunmap_local(first_entry);
473		brelse(entry_bh);
474	}
475
476	return nvi;
477}
478
479/**
480 * nilfs_dat_read - read or get dat inode
481 * @sb: super block instance
482 * @entry_size: size of a dat entry
483 * @raw_inode: on-disk dat inode
484 * @inodep: buffer to store the inode
485 *
486 * Return: 0 on success, or a negative error code on failure.
487 */
488int nilfs_dat_read(struct super_block *sb, size_t entry_size,
489		   struct nilfs_inode *raw_inode, struct inode **inodep)
490{
491	static struct lock_class_key dat_lock_key;
492	struct inode *dat;
493	struct nilfs_dat_info *di;
494	int err;
495
496	if (entry_size > sb->s_blocksize) {
497		nilfs_err(sb, "too large DAT entry size: %zu bytes",
498			  entry_size);
499		return -EINVAL;
500	} else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) {
501		nilfs_err(sb, "too small DAT entry size: %zu bytes",
502			  entry_size);
503		return -EINVAL;
504	}
505
506	dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
507	if (unlikely(!dat))
508		return -ENOMEM;
509	if (!(inode_state_read_once(dat) & I_NEW))
510		goto out;
511
512	err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
513	if (err)
514		goto failed;
515
516	err = nilfs_palloc_init_blockgroup(dat, entry_size);
517	if (err)
518		goto failed;
519
520	di = NILFS_DAT_I(dat);
521	lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
522	nilfs_palloc_setup_cache(dat, &di->palloc_cache);
523	err = nilfs_mdt_setup_shadow_map(dat, &di->shadow);
524	if (err)
525		goto failed;
526
527	err = nilfs_attach_btree_node_cache(dat);
528	if (err)
529		goto failed;
530	err = nilfs_read_inode_common(dat, raw_inode);
531	if (err)
532		goto failed;
533
534	unlock_new_inode(dat);
535 out:
536	*inodep = dat;
537	return 0;
538 failed:
539	iget_failed(dat);
540	return err;
541}