1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * NILFS segment usage file.
4 *
5 * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Written by Koji Sato.
8 * Revised by Ryusuke Konishi.
9 */
10
11#include <linux/kernel.h>
12#include <linux/fs.h>
13#include <linux/string.h>
14#include <linux/buffer_head.h>
15#include <linux/errno.h>
16#include "mdt.h"
17#include "sufile.h"
18
19#include <trace/events/nilfs2.h>
20
21/**
22 * struct nilfs_sufile_info - on-memory private data of sufile
23 * @mi: on-memory private data of metadata file
24 * @ncleansegs: number of clean segments
25 * @allocmin: lower limit of allocatable segment range
26 * @allocmax: upper limit of allocatable segment range
27 */
28struct nilfs_sufile_info {
29	struct nilfs_mdt_info mi;
30	unsigned long ncleansegs;/* number of clean segments */
31	__u64 allocmin;		/* lower limit of allocatable segment range */
32	__u64 allocmax;		/* upper limit of allocatable segment range */
33};
34
35static inline struct nilfs_sufile_info *NILFS_SUI(struct inode *sufile)
36{
37	return (struct nilfs_sufile_info *)NILFS_MDT(sufile);
38}
39
40static inline unsigned long
41nilfs_sufile_segment_usages_per_block(const struct inode *sufile)
42{
43	return NILFS_MDT(sufile)->mi_entries_per_block;
44}
45
46static unsigned long
47nilfs_sufile_get_blkoff(const struct inode *sufile, __u64 segnum)
48{
49	__u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
50
51	t = div64_ul(t, nilfs_sufile_segment_usages_per_block(sufile));
52	return (unsigned long)t;
53}
54
55static unsigned long
56nilfs_sufile_get_offset(const struct inode *sufile, __u64 segnum)
57{
58	__u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
59
60	return do_div(t, nilfs_sufile_segment_usages_per_block(sufile));
61}
62
63static unsigned long
64nilfs_sufile_segment_usages_in_block(const struct inode *sufile, __u64 curr,
65				     __u64 max)
66{
67	return min_t(unsigned long,
68		     nilfs_sufile_segment_usages_per_block(sufile) -
69		     nilfs_sufile_get_offset(sufile, curr),
70		     max - curr + 1);
71}
72
73static struct nilfs_segment_usage *
74nilfs_sufile_block_get_segment_usage(const struct inode *sufile, __u64 segnum,
75				     struct buffer_head *bh, void *kaddr)
76{
77	return kaddr + bh_offset(bh) +
78		nilfs_sufile_get_offset(sufile, segnum) *
79		NILFS_MDT(sufile)->mi_entry_size;
80}
81
82static inline int nilfs_sufile_get_header_block(struct inode *sufile,
83						struct buffer_head **bhp)
84{
85	return nilfs_mdt_get_block(sufile, 0, 0, NULL, bhp);
86}
87
88static inline int
89nilfs_sufile_get_segment_usage_block(struct inode *sufile, __u64 segnum,
90				     int create, struct buffer_head **bhp)
91{
92	return nilfs_mdt_get_block(sufile,
93				   nilfs_sufile_get_blkoff(sufile, segnum),
94				   create, NULL, bhp);
95}
96
97static int nilfs_sufile_delete_segment_usage_block(struct inode *sufile,
98						   __u64 segnum)
99{
100	return nilfs_mdt_delete_block(sufile,
101				      nilfs_sufile_get_blkoff(sufile, segnum));
102}
103
104static void nilfs_sufile_mod_counter(struct buffer_head *header_bh,
105				     u64 ncleanadd, u64 ndirtyadd)
106{
107	struct nilfs_sufile_header *header;
108	void *kaddr;
109
110	kaddr = kmap_local_page(header_bh->b_page);
111	header = kaddr + bh_offset(header_bh);
112	le64_add_cpu(&header->sh_ncleansegs, ncleanadd);
113	le64_add_cpu(&header->sh_ndirtysegs, ndirtyadd);
114	kunmap_local(kaddr);
115
116	mark_buffer_dirty(header_bh);
117}
118
119/**
120 * nilfs_sufile_get_ncleansegs - return the number of clean segments
121 * @sufile: inode of segment usage file
122 */
123unsigned long nilfs_sufile_get_ncleansegs(struct inode *sufile)
124{
125	return NILFS_SUI(sufile)->ncleansegs;
126}
127
128/**
129 * nilfs_sufile_updatev - modify multiple segment usages at a time
130 * @sufile: inode of segment usage file
131 * @segnumv: array of segment numbers
132 * @nsegs: size of @segnumv array
133 * @create: creation flag
134 * @ndone: place to store number of modified segments on @segnumv
135 * @dofunc: primitive operation for the update
136 *
137 * Description: nilfs_sufile_updatev() repeatedly calls @dofunc
138 * against the given array of segments.  The @dofunc is called with
139 * buffers of a header block and the sufile block in which the target
140 * segment usage entry is contained.  If @ndone is given, the number
141 * of successfully modified segments from the head is stored in the
142 * place @ndone points to.
143 *
144 * Return Value: On success, zero is returned.  On error, one of the
145 * following negative error codes is returned.
146 *
147 * %-EIO - I/O error.
148 *
149 * %-ENOMEM - Insufficient amount of memory available.
150 *
151 * %-ENOENT - Given segment usage is in hole block (may be returned if
152 *            @create is zero)
153 *
154 * %-EINVAL - Invalid segment usage number
155 */
156int nilfs_sufile_updatev(struct inode *sufile, __u64 *segnumv, size_t nsegs,
157			 int create, size_t *ndone,
158			 void (*dofunc)(struct inode *, __u64,
159					struct buffer_head *,
160					struct buffer_head *))
161{
162	struct buffer_head *header_bh, *bh;
163	unsigned long blkoff, prev_blkoff;
164	__u64 *seg;
165	size_t nerr = 0, n = 0;
166	int ret = 0;
167
168	if (unlikely(nsegs == 0))
169		goto out;
170
171	down_write(&NILFS_MDT(sufile)->mi_sem);
172	for (seg = segnumv; seg < segnumv + nsegs; seg++) {
173		if (unlikely(*seg >= nilfs_sufile_get_nsegments(sufile))) {
174			nilfs_warn(sufile->i_sb,
175				   "%s: invalid segment number: %llu",
176				   __func__, (unsigned long long)*seg);
177			nerr++;
178		}
179	}
180	if (nerr > 0) {
181		ret = -EINVAL;
182		goto out_sem;
183	}
184
185	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
186	if (ret < 0)
187		goto out_sem;
188
189	seg = segnumv;
190	blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
191	ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
192	if (ret < 0)
193		goto out_header;
194
195	for (;;) {
196		dofunc(sufile, *seg, header_bh, bh);
197
198		if (++seg >= segnumv + nsegs)
199			break;
200		prev_blkoff = blkoff;
201		blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
202		if (blkoff == prev_blkoff)
203			continue;
204
205		/* get different block */
206		brelse(bh);
207		ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
208		if (unlikely(ret < 0))
209			goto out_header;
210	}
211	brelse(bh);
212
213 out_header:
214	n = seg - segnumv;
215	brelse(header_bh);
216 out_sem:
217	up_write(&NILFS_MDT(sufile)->mi_sem);
218 out:
219	if (ndone)
220		*ndone = n;
221	return ret;
222}
223
224int nilfs_sufile_update(struct inode *sufile, __u64 segnum, int create,
225			void (*dofunc)(struct inode *, __u64,
226				       struct buffer_head *,
227				       struct buffer_head *))
228{
229	struct buffer_head *header_bh, *bh;
230	int ret;
231
232	if (unlikely(segnum >= nilfs_sufile_get_nsegments(sufile))) {
233		nilfs_warn(sufile->i_sb, "%s: invalid segment number: %llu",
234			   __func__, (unsigned long long)segnum);
235		return -EINVAL;
236	}
237	down_write(&NILFS_MDT(sufile)->mi_sem);
238
239	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
240	if (ret < 0)
241		goto out_sem;
242
243	ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, create, &bh);
244	if (!ret) {
245		dofunc(sufile, segnum, header_bh, bh);
246		brelse(bh);
247	}
248	brelse(header_bh);
249
250 out_sem:
251	up_write(&NILFS_MDT(sufile)->mi_sem);
252	return ret;
253}
254
255/**
256 * nilfs_sufile_set_alloc_range - limit range of segment to be allocated
257 * @sufile: inode of segment usage file
258 * @start: minimum segment number of allocatable region (inclusive)
259 * @end: maximum segment number of allocatable region (inclusive)
260 *
261 * Return Value: On success, 0 is returned.  On error, one of the
262 * following negative error codes is returned.
263 *
264 * %-ERANGE - invalid segment region
265 */
266int nilfs_sufile_set_alloc_range(struct inode *sufile, __u64 start, __u64 end)
267{
268	struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
269	__u64 nsegs;
270	int ret = -ERANGE;
271
272	down_write(&NILFS_MDT(sufile)->mi_sem);
273	nsegs = nilfs_sufile_get_nsegments(sufile);
274
275	if (start <= end && end < nsegs) {
276		sui->allocmin = start;
277		sui->allocmax = end;
278		ret = 0;
279	}
280	up_write(&NILFS_MDT(sufile)->mi_sem);
281	return ret;
282}
283
284/**
285 * nilfs_sufile_alloc - allocate a segment
286 * @sufile: inode of segment usage file
287 * @segnump: pointer to segment number
288 *
289 * Description: nilfs_sufile_alloc() allocates a clean segment.
290 *
291 * Return Value: On success, 0 is returned and the segment number of the
292 * allocated segment is stored in the place pointed by @segnump. On error, one
293 * of the following negative error codes is returned.
294 *
295 * %-EIO - I/O error.
296 *
297 * %-ENOMEM - Insufficient amount of memory available.
298 *
299 * %-ENOSPC - No clean segment left.
300 */
301int nilfs_sufile_alloc(struct inode *sufile, __u64 *segnump)
302{
303	struct buffer_head *header_bh, *su_bh;
304	struct nilfs_sufile_header *header;
305	struct nilfs_segment_usage *su;
306	struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
307	size_t susz = NILFS_MDT(sufile)->mi_entry_size;
308	__u64 segnum, maxsegnum, last_alloc;
309	void *kaddr;
310	unsigned long nsegments, nsus, cnt;
311	int ret, j;
312
313	down_write(&NILFS_MDT(sufile)->mi_sem);
314
315	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
316	if (ret < 0)
317		goto out_sem;
318	kaddr = kmap_local_page(header_bh->b_page);
319	header = kaddr + bh_offset(header_bh);
320	last_alloc = le64_to_cpu(header->sh_last_alloc);
321	kunmap_local(kaddr);
322
323	nsegments = nilfs_sufile_get_nsegments(sufile);
324	maxsegnum = sui->allocmax;
325	segnum = last_alloc + 1;
326	if (segnum < sui->allocmin || segnum > sui->allocmax)
327		segnum = sui->allocmin;
328
329	for (cnt = 0; cnt < nsegments; cnt += nsus) {
330		if (segnum > maxsegnum) {
331			if (cnt < sui->allocmax - sui->allocmin + 1) {
332				/*
333				 * wrap around in the limited region.
334				 * if allocation started from
335				 * sui->allocmin, this never happens.
336				 */
337				segnum = sui->allocmin;
338				maxsegnum = last_alloc;
339			} else if (segnum > sui->allocmin &&
340				   sui->allocmax + 1 < nsegments) {
341				segnum = sui->allocmax + 1;
342				maxsegnum = nsegments - 1;
343			} else if (sui->allocmin > 0)  {
344				segnum = 0;
345				maxsegnum = sui->allocmin - 1;
346			} else {
347				break; /* never happens */
348			}
349		}
350		trace_nilfs2_segment_usage_check(sufile, segnum, cnt);
351		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 1,
352							   &su_bh);
353		if (ret < 0)
354			goto out_header;
355		kaddr = kmap_local_page(su_bh->b_page);
356		su = nilfs_sufile_block_get_segment_usage(
357			sufile, segnum, su_bh, kaddr);
358
359		nsus = nilfs_sufile_segment_usages_in_block(
360			sufile, segnum, maxsegnum);
361		for (j = 0; j < nsus; j++, su = (void *)su + susz, segnum++) {
362			if (!nilfs_segment_usage_clean(su))
363				continue;
364			/* found a clean segment */
365			nilfs_segment_usage_set_dirty(su);
366			kunmap_local(kaddr);
367
368			kaddr = kmap_local_page(header_bh->b_page);
369			header = kaddr + bh_offset(header_bh);
370			le64_add_cpu(&header->sh_ncleansegs, -1);
371			le64_add_cpu(&header->sh_ndirtysegs, 1);
372			header->sh_last_alloc = cpu_to_le64(segnum);
373			kunmap_local(kaddr);
374
375			sui->ncleansegs--;
376			mark_buffer_dirty(header_bh);
377			mark_buffer_dirty(su_bh);
378			nilfs_mdt_mark_dirty(sufile);
379			brelse(su_bh);
380			*segnump = segnum;
381
382			trace_nilfs2_segment_usage_allocated(sufile, segnum);
383
384			goto out_header;
385		}
386
387		kunmap_local(kaddr);
388		brelse(su_bh);
389	}
390
391	/* no segments left */
392	ret = -ENOSPC;
393
394 out_header:
395	brelse(header_bh);
396
397 out_sem:
398	up_write(&NILFS_MDT(sufile)->mi_sem);
399	return ret;
400}
401
402void nilfs_sufile_do_cancel_free(struct inode *sufile, __u64 segnum,
403				 struct buffer_head *header_bh,
404				 struct buffer_head *su_bh)
405{
406	struct nilfs_segment_usage *su;
407	void *kaddr;
408
409	kaddr = kmap_local_page(su_bh->b_page);
410	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
411	if (unlikely(!nilfs_segment_usage_clean(su))) {
412		nilfs_warn(sufile->i_sb, "%s: segment %llu must be clean",
413			   __func__, (unsigned long long)segnum);
414		kunmap_local(kaddr);
415		return;
416	}
417	nilfs_segment_usage_set_dirty(su);
418	kunmap_local(kaddr);
419
420	nilfs_sufile_mod_counter(header_bh, -1, 1);
421	NILFS_SUI(sufile)->ncleansegs--;
422
423	mark_buffer_dirty(su_bh);
424	nilfs_mdt_mark_dirty(sufile);
425}
426
427void nilfs_sufile_do_scrap(struct inode *sufile, __u64 segnum,
428			   struct buffer_head *header_bh,
429			   struct buffer_head *su_bh)
430{
431	struct nilfs_segment_usage *su;
432	void *kaddr;
433	int clean, dirty;
434
435	kaddr = kmap_local_page(su_bh->b_page);
436	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
437	if (su->su_flags == cpu_to_le32(BIT(NILFS_SEGMENT_USAGE_DIRTY)) &&
438	    su->su_nblocks == cpu_to_le32(0)) {
439		kunmap_local(kaddr);
440		return;
441	}
442	clean = nilfs_segment_usage_clean(su);
443	dirty = nilfs_segment_usage_dirty(su);
444
445	/* make the segment garbage */
446	su->su_lastmod = cpu_to_le64(0);
447	su->su_nblocks = cpu_to_le32(0);
448	su->su_flags = cpu_to_le32(BIT(NILFS_SEGMENT_USAGE_DIRTY));
449	kunmap_local(kaddr);
450
451	nilfs_sufile_mod_counter(header_bh, clean ? (u64)-1 : 0, dirty ? 0 : 1);
452	NILFS_SUI(sufile)->ncleansegs -= clean;
453
454	mark_buffer_dirty(su_bh);
455	nilfs_mdt_mark_dirty(sufile);
456}
457
458void nilfs_sufile_do_free(struct inode *sufile, __u64 segnum,
459			  struct buffer_head *header_bh,
460			  struct buffer_head *su_bh)
461{
462	struct nilfs_segment_usage *su;
463	void *kaddr;
464	int sudirty;
465
466	kaddr = kmap_local_page(su_bh->b_page);
467	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
468	if (nilfs_segment_usage_clean(su)) {
469		nilfs_warn(sufile->i_sb, "%s: segment %llu is already clean",
470			   __func__, (unsigned long long)segnum);
471		kunmap_local(kaddr);
472		return;
473	}
474	if (unlikely(nilfs_segment_usage_error(su)))
475		nilfs_warn(sufile->i_sb, "free segment %llu marked in error",
476			   (unsigned long long)segnum);
477
478	sudirty = nilfs_segment_usage_dirty(su);
479	if (unlikely(!sudirty))
480		nilfs_warn(sufile->i_sb, "free unallocated segment %llu",
481			   (unsigned long long)segnum);
482
483	nilfs_segment_usage_set_clean(su);
484	kunmap_local(kaddr);
485	mark_buffer_dirty(su_bh);
486
487	nilfs_sufile_mod_counter(header_bh, 1, sudirty ? (u64)-1 : 0);
488	NILFS_SUI(sufile)->ncleansegs++;
489
490	nilfs_mdt_mark_dirty(sufile);
491
492	trace_nilfs2_segment_usage_freed(sufile, segnum);
493}
494
495/**
496 * nilfs_sufile_mark_dirty - mark the buffer having a segment usage dirty
497 * @sufile: inode of segment usage file
498 * @segnum: segment number
499 */
500int nilfs_sufile_mark_dirty(struct inode *sufile, __u64 segnum)
501{
502	struct buffer_head *bh;
503	void *kaddr;
504	struct nilfs_segment_usage *su;
505	int ret;
506
507	down_write(&NILFS_MDT(sufile)->mi_sem);
508	ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
509	if (ret)
510		goto out_sem;
511
512	kaddr = kmap_local_page(bh->b_page);
513	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);
514	if (unlikely(nilfs_segment_usage_error(su))) {
515		struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
516
517		kunmap_local(kaddr);
518		brelse(bh);
519		if (nilfs_segment_is_active(nilfs, segnum)) {
520			nilfs_error(sufile->i_sb,
521				    "active segment %llu is erroneous",
522				    (unsigned long long)segnum);
523		} else {
524			/*
525			 * Segments marked erroneous are never allocated by
526			 * nilfs_sufile_alloc(); only active segments, ie,
527			 * the segments indexed by ns_segnum or ns_nextnum,
528			 * can be erroneous here.
529			 */
530			WARN_ON_ONCE(1);
531		}
532		ret = -EIO;
533	} else {
534		nilfs_segment_usage_set_dirty(su);
535		kunmap_local(kaddr);
536		mark_buffer_dirty(bh);
537		nilfs_mdt_mark_dirty(sufile);
538		brelse(bh);
539	}
540out_sem:
541	up_write(&NILFS_MDT(sufile)->mi_sem);
542	return ret;
543}
544
545/**
546 * nilfs_sufile_set_segment_usage - set usage of a segment
547 * @sufile: inode of segment usage file
548 * @segnum: segment number
549 * @nblocks: number of live blocks in the segment
550 * @modtime: modification time (option)
551 */
552int nilfs_sufile_set_segment_usage(struct inode *sufile, __u64 segnum,
553				   unsigned long nblocks, time64_t modtime)
554{
555	struct buffer_head *bh;
556	struct nilfs_segment_usage *su;
557	void *kaddr;
558	int ret;
559
560	down_write(&NILFS_MDT(sufile)->mi_sem);
561	ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
562	if (ret < 0)
563		goto out_sem;
564
565	kaddr = kmap_local_page(bh->b_page);
566	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);
567	if (modtime) {
568		/*
569		 * Check segusage error and set su_lastmod only when updating
570		 * this entry with a valid timestamp, not for cancellation.
571		 */
572		WARN_ON_ONCE(nilfs_segment_usage_error(su));
573		su->su_lastmod = cpu_to_le64(modtime);
574	}
575	su->su_nblocks = cpu_to_le32(nblocks);
576	kunmap_local(kaddr);
577
578	mark_buffer_dirty(bh);
579	nilfs_mdt_mark_dirty(sufile);
580	brelse(bh);
581
582 out_sem:
583	up_write(&NILFS_MDT(sufile)->mi_sem);
584	return ret;
585}
586
587/**
588 * nilfs_sufile_get_stat - get segment usage statistics
589 * @sufile: inode of segment usage file
590 * @sustat: pointer to a structure of segment usage statistics
591 *
592 * Description: nilfs_sufile_get_stat() returns information about segment
593 * usage.
594 *
595 * Return Value: On success, 0 is returned, and segment usage information is
596 * stored in the place pointed by @sustat. On error, one of the following
597 * negative error codes is returned.
598 *
599 * %-EIO - I/O error.
600 *
601 * %-ENOMEM - Insufficient amount of memory available.
602 */
603int nilfs_sufile_get_stat(struct inode *sufile, struct nilfs_sustat *sustat)
604{
605	struct buffer_head *header_bh;
606	struct nilfs_sufile_header *header;
607	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
608	void *kaddr;
609	int ret;
610
611	down_read(&NILFS_MDT(sufile)->mi_sem);
612
613	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
614	if (ret < 0)
615		goto out_sem;
616
617	kaddr = kmap_local_page(header_bh->b_page);
618	header = kaddr + bh_offset(header_bh);
619	sustat->ss_nsegs = nilfs_sufile_get_nsegments(sufile);
620	sustat->ss_ncleansegs = le64_to_cpu(header->sh_ncleansegs);
621	sustat->ss_ndirtysegs = le64_to_cpu(header->sh_ndirtysegs);
622	sustat->ss_ctime = nilfs->ns_ctime;
623	sustat->ss_nongc_ctime = nilfs->ns_nongc_ctime;
624	spin_lock(&nilfs->ns_last_segment_lock);
625	sustat->ss_prot_seq = nilfs->ns_prot_seq;
626	spin_unlock(&nilfs->ns_last_segment_lock);
627	kunmap_local(kaddr);
628	brelse(header_bh);
629
630 out_sem:
631	up_read(&NILFS_MDT(sufile)->mi_sem);
632	return ret;
633}
634
635void nilfs_sufile_do_set_error(struct inode *sufile, __u64 segnum,
636			       struct buffer_head *header_bh,
637			       struct buffer_head *su_bh)
638{
639	struct nilfs_segment_usage *su;
640	void *kaddr;
641	int suclean;
642
643	kaddr = kmap_local_page(su_bh->b_page);
644	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
645	if (nilfs_segment_usage_error(su)) {
646		kunmap_local(kaddr);
647		return;
648	}
649	suclean = nilfs_segment_usage_clean(su);
650	nilfs_segment_usage_set_error(su);
651	kunmap_local(kaddr);
652
653	if (suclean) {
654		nilfs_sufile_mod_counter(header_bh, -1, 0);
655		NILFS_SUI(sufile)->ncleansegs--;
656	}
657	mark_buffer_dirty(su_bh);
658	nilfs_mdt_mark_dirty(sufile);
659}
660
661/**
662 * nilfs_sufile_truncate_range - truncate range of segment array
663 * @sufile: inode of segment usage file
664 * @start: start segment number (inclusive)
665 * @end: end segment number (inclusive)
666 *
667 * Return Value: On success, 0 is returned.  On error, one of the
668 * following negative error codes is returned.
669 *
670 * %-EIO - I/O error.
671 *
672 * %-ENOMEM - Insufficient amount of memory available.
673 *
674 * %-EINVAL - Invalid number of segments specified
675 *
676 * %-EBUSY - Dirty or active segments are present in the range
677 */
678static int nilfs_sufile_truncate_range(struct inode *sufile,
679				       __u64 start, __u64 end)
680{
681	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
682	struct buffer_head *header_bh;
683	struct buffer_head *su_bh;
684	struct nilfs_segment_usage *su, *su2;
685	size_t susz = NILFS_MDT(sufile)->mi_entry_size;
686	unsigned long segusages_per_block;
687	unsigned long nsegs, ncleaned;
688	__u64 segnum;
689	void *kaddr;
690	ssize_t n, nc;
691	int ret;
692	int j;
693
694	nsegs = nilfs_sufile_get_nsegments(sufile);
695
696	ret = -EINVAL;
697	if (start > end || start >= nsegs)
698		goto out;
699
700	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
701	if (ret < 0)
702		goto out;
703
704	segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
705	ncleaned = 0;
706
707	for (segnum = start; segnum <= end; segnum += n) {
708		n = min_t(unsigned long,
709			  segusages_per_block -
710				  nilfs_sufile_get_offset(sufile, segnum),
711			  end - segnum + 1);
712		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
713							   &su_bh);
714		if (ret < 0) {
715			if (ret != -ENOENT)
716				goto out_header;
717			/* hole */
718			continue;
719		}
720		kaddr = kmap_local_page(su_bh->b_page);
721		su = nilfs_sufile_block_get_segment_usage(
722			sufile, segnum, su_bh, kaddr);
723		su2 = su;
724		for (j = 0; j < n; j++, su = (void *)su + susz) {
725			if ((le32_to_cpu(su->su_flags) &
726			     ~BIT(NILFS_SEGMENT_USAGE_ERROR)) ||
727			    nilfs_segment_is_active(nilfs, segnum + j)) {
728				ret = -EBUSY;
729				kunmap_local(kaddr);
730				brelse(su_bh);
731				goto out_header;
732			}
733		}
734		nc = 0;
735		for (su = su2, j = 0; j < n; j++, su = (void *)su + susz) {
736			if (nilfs_segment_usage_error(su)) {
737				nilfs_segment_usage_set_clean(su);
738				nc++;
739			}
740		}
741		kunmap_local(kaddr);
742		if (nc > 0) {
743			mark_buffer_dirty(su_bh);
744			ncleaned += nc;
745		}
746		brelse(su_bh);
747
748		if (n == segusages_per_block) {
749			/* make hole */
750			nilfs_sufile_delete_segment_usage_block(sufile, segnum);
751		}
752	}
753	ret = 0;
754
755out_header:
756	if (ncleaned > 0) {
757		NILFS_SUI(sufile)->ncleansegs += ncleaned;
758		nilfs_sufile_mod_counter(header_bh, ncleaned, 0);
759		nilfs_mdt_mark_dirty(sufile);
760	}
761	brelse(header_bh);
762out:
763	return ret;
764}
765
766/**
767 * nilfs_sufile_resize - resize segment array
768 * @sufile: inode of segment usage file
769 * @newnsegs: new number of segments
770 *
771 * Return Value: On success, 0 is returned.  On error, one of the
772 * following negative error codes is returned.
773 *
774 * %-EIO - I/O error.
775 *
776 * %-ENOMEM - Insufficient amount of memory available.
777 *
778 * %-ENOSPC - Enough free space is not left for shrinking
779 *
780 * %-EBUSY - Dirty or active segments exist in the region to be truncated
781 */
782int nilfs_sufile_resize(struct inode *sufile, __u64 newnsegs)
783{
784	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
785	struct buffer_head *header_bh;
786	struct nilfs_sufile_header *header;
787	struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
788	void *kaddr;
789	unsigned long nsegs, nrsvsegs;
790	int ret = 0;
791
792	down_write(&NILFS_MDT(sufile)->mi_sem);
793
794	nsegs = nilfs_sufile_get_nsegments(sufile);
795	if (nsegs == newnsegs)
796		goto out;
797
798	ret = -ENOSPC;
799	nrsvsegs = nilfs_nrsvsegs(nilfs, newnsegs);
800	if (newnsegs < nsegs && nsegs - newnsegs + nrsvsegs > sui->ncleansegs)
801		goto out;
802
803	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
804	if (ret < 0)
805		goto out;
806
807	if (newnsegs > nsegs) {
808		sui->ncleansegs += newnsegs - nsegs;
809	} else /* newnsegs < nsegs */ {
810		ret = nilfs_sufile_truncate_range(sufile, newnsegs, nsegs - 1);
811		if (ret < 0)
812			goto out_header;
813
814		sui->ncleansegs -= nsegs - newnsegs;
815
816		/*
817		 * If the sufile is successfully truncated, immediately adjust
818		 * the segment allocation space while locking the semaphore
819		 * "mi_sem" so that nilfs_sufile_alloc() never allocates
820		 * segments in the truncated space.
821		 */
822		sui->allocmax = newnsegs - 1;
823		sui->allocmin = 0;
824	}
825
826	kaddr = kmap_local_page(header_bh->b_page);
827	header = kaddr + bh_offset(header_bh);
828	header->sh_ncleansegs = cpu_to_le64(sui->ncleansegs);
829	kunmap_local(kaddr);
830
831	mark_buffer_dirty(header_bh);
832	nilfs_mdt_mark_dirty(sufile);
833	nilfs_set_nsegments(nilfs, newnsegs);
834
835out_header:
836	brelse(header_bh);
837out:
838	up_write(&NILFS_MDT(sufile)->mi_sem);
839	return ret;
840}
841
842/**
843 * nilfs_sufile_get_suinfo -
844 * @sufile: inode of segment usage file
845 * @segnum: segment number to start looking
846 * @buf: array of suinfo
847 * @sisz: byte size of suinfo
848 * @nsi: size of suinfo array
849 *
850 * Description:
851 *
852 * Return Value: On success, 0 is returned and .... On error, one of the
853 * following negative error codes is returned.
854 *
855 * %-EIO - I/O error.
856 *
857 * %-ENOMEM - Insufficient amount of memory available.
858 */
859ssize_t nilfs_sufile_get_suinfo(struct inode *sufile, __u64 segnum, void *buf,
860				unsigned int sisz, size_t nsi)
861{
862	struct buffer_head *su_bh;
863	struct nilfs_segment_usage *su;
864	struct nilfs_suinfo *si = buf;
865	size_t susz = NILFS_MDT(sufile)->mi_entry_size;
866	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
867	void *kaddr;
868	unsigned long nsegs, segusages_per_block;
869	ssize_t n;
870	int ret, i, j;
871
872	down_read(&NILFS_MDT(sufile)->mi_sem);
873
874	segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
875	nsegs = min_t(unsigned long,
876		      nilfs_sufile_get_nsegments(sufile) - segnum,
877		      nsi);
878	for (i = 0; i < nsegs; i += n, segnum += n) {
879		n = min_t(unsigned long,
880			  segusages_per_block -
881				  nilfs_sufile_get_offset(sufile, segnum),
882			  nsegs - i);
883		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
884							   &su_bh);
885		if (ret < 0) {
886			if (ret != -ENOENT)
887				goto out;
888			/* hole */
889			memset(si, 0, sisz * n);
890			si = (void *)si + sisz * n;
891			continue;
892		}
893
894		kaddr = kmap_local_page(su_bh->b_page);
895		su = nilfs_sufile_block_get_segment_usage(
896			sufile, segnum, su_bh, kaddr);
897		for (j = 0; j < n;
898		     j++, su = (void *)su + susz, si = (void *)si + sisz) {
899			si->sui_lastmod = le64_to_cpu(su->su_lastmod);
900			si->sui_nblocks = le32_to_cpu(su->su_nblocks);
901			si->sui_flags = le32_to_cpu(su->su_flags) &
902				~BIT(NILFS_SEGMENT_USAGE_ACTIVE);
903			if (nilfs_segment_is_active(nilfs, segnum + j))
904				si->sui_flags |=
905					BIT(NILFS_SEGMENT_USAGE_ACTIVE);
906		}
907		kunmap_local(kaddr);
908		brelse(su_bh);
909	}
910	ret = nsegs;
911
912 out:
913	up_read(&NILFS_MDT(sufile)->mi_sem);
914	return ret;
915}
916
917/**
918 * nilfs_sufile_set_suinfo - sets segment usage info
919 * @sufile: inode of segment usage file
920 * @buf: array of suinfo_update
921 * @supsz: byte size of suinfo_update
922 * @nsup: size of suinfo_update array
923 *
924 * Description: Takes an array of nilfs_suinfo_update structs and updates
925 * segment usage accordingly. Only the fields indicated by the sup_flags
926 * are updated.
927 *
928 * Return Value: On success, 0 is returned. On error, one of the
929 * following negative error codes is returned.
930 *
931 * %-EIO - I/O error.
932 *
933 * %-ENOMEM - Insufficient amount of memory available.
934 *
935 * %-EINVAL - Invalid values in input (segment number, flags or nblocks)
936 */
937ssize_t nilfs_sufile_set_suinfo(struct inode *sufile, void *buf,
938				unsigned int supsz, size_t nsup)
939{
940	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
941	struct buffer_head *header_bh, *bh;
942	struct nilfs_suinfo_update *sup, *supend = buf + supsz * nsup;
943	struct nilfs_segment_usage *su;
944	void *kaddr;
945	unsigned long blkoff, prev_blkoff;
946	int cleansi, cleansu, dirtysi, dirtysu;
947	long ncleaned = 0, ndirtied = 0;
948	int ret = 0;
949
950	if (unlikely(nsup == 0))
951		return ret;
952
953	for (sup = buf; sup < supend; sup = (void *)sup + supsz) {
954		if (sup->sup_segnum >= nilfs->ns_nsegments
955			|| (sup->sup_flags &
956				(~0UL << __NR_NILFS_SUINFO_UPDATE_FIELDS))
957			|| (nilfs_suinfo_update_nblocks(sup) &&
958				sup->sup_sui.sui_nblocks >
959				nilfs->ns_blocks_per_segment))
960			return -EINVAL;
961	}
962
963	down_write(&NILFS_MDT(sufile)->mi_sem);
964
965	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
966	if (ret < 0)
967		goto out_sem;
968
969	sup = buf;
970	blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
971	ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
972	if (ret < 0)
973		goto out_header;
974
975	for (;;) {
976		kaddr = kmap_local_page(bh->b_page);
977		su = nilfs_sufile_block_get_segment_usage(
978			sufile, sup->sup_segnum, bh, kaddr);
979
980		if (nilfs_suinfo_update_lastmod(sup))
981			su->su_lastmod = cpu_to_le64(sup->sup_sui.sui_lastmod);
982
983		if (nilfs_suinfo_update_nblocks(sup))
984			su->su_nblocks = cpu_to_le32(sup->sup_sui.sui_nblocks);
985
986		if (nilfs_suinfo_update_flags(sup)) {
987			/*
988			 * Active flag is a virtual flag projected by running
989			 * nilfs kernel code - drop it not to write it to
990			 * disk.
991			 */
992			sup->sup_sui.sui_flags &=
993					~BIT(NILFS_SEGMENT_USAGE_ACTIVE);
994
995			cleansi = nilfs_suinfo_clean(&sup->sup_sui);
996			cleansu = nilfs_segment_usage_clean(su);
997			dirtysi = nilfs_suinfo_dirty(&sup->sup_sui);
998			dirtysu = nilfs_segment_usage_dirty(su);
999
1000			if (cleansi && !cleansu)
1001				++ncleaned;
1002			else if (!cleansi && cleansu)
1003				--ncleaned;
1004
1005			if (dirtysi && !dirtysu)
1006				++ndirtied;
1007			else if (!dirtysi && dirtysu)
1008				--ndirtied;
1009
1010			su->su_flags = cpu_to_le32(sup->sup_sui.sui_flags);
1011		}
1012
1013		kunmap_local(kaddr);
1014
1015		sup = (void *)sup + supsz;
1016		if (sup >= supend)
1017			break;
1018
1019		prev_blkoff = blkoff;
1020		blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
1021		if (blkoff == prev_blkoff)
1022			continue;
1023
1024		/* get different block */
1025		mark_buffer_dirty(bh);
1026		put_bh(bh);
1027		ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
1028		if (unlikely(ret < 0))
1029			goto out_mark;
1030	}
1031	mark_buffer_dirty(bh);
1032	put_bh(bh);
1033
1034 out_mark:
1035	if (ncleaned || ndirtied) {
1036		nilfs_sufile_mod_counter(header_bh, (u64)ncleaned,
1037				(u64)ndirtied);
1038		NILFS_SUI(sufile)->ncleansegs += ncleaned;
1039	}
1040	nilfs_mdt_mark_dirty(sufile);
1041 out_header:
1042	put_bh(header_bh);
1043 out_sem:
1044	up_write(&NILFS_MDT(sufile)->mi_sem);
1045	return ret;
1046}
1047
1048/**
1049 * nilfs_sufile_trim_fs() - trim ioctl handle function
1050 * @sufile: inode of segment usage file
1051 * @range: fstrim_range structure
1052 *
1053 * start:	First Byte to trim
1054 * len:		number of Bytes to trim from start
1055 * minlen:	minimum extent length in Bytes
1056 *
1057 * Decription: nilfs_sufile_trim_fs goes through all segments containing bytes
1058 * from start to start+len. start is rounded up to the next block boundary
1059 * and start+len is rounded down. For each clean segment blkdev_issue_discard
1060 * function is invoked.
1061 *
1062 * Return Value: On success, 0 is returned or negative error code, otherwise.
1063 */
1064int nilfs_sufile_trim_fs(struct inode *sufile, struct fstrim_range *range)
1065{
1066	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
1067	struct buffer_head *su_bh;
1068	struct nilfs_segment_usage *su;
1069	void *kaddr;
1070	size_t n, i, susz = NILFS_MDT(sufile)->mi_entry_size;
1071	sector_t seg_start, seg_end, start_block, end_block;
1072	sector_t start = 0, nblocks = 0;
1073	u64 segnum, segnum_end, minlen, len, max_blocks, ndiscarded = 0;
1074	int ret = 0;
1075	unsigned int sects_per_block;
1076
1077	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
1078			bdev_logical_block_size(nilfs->ns_bdev);
1079	len = range->len >> nilfs->ns_blocksize_bits;
1080	minlen = range->minlen >> nilfs->ns_blocksize_bits;
1081	max_blocks = ((u64)nilfs->ns_nsegments * nilfs->ns_blocks_per_segment);
1082
1083	if (!len || range->start >= max_blocks << nilfs->ns_blocksize_bits)
1084		return -EINVAL;
1085
1086	start_block = (range->start + nilfs->ns_blocksize - 1) >>
1087			nilfs->ns_blocksize_bits;
1088
1089	/*
1090	 * range->len can be very large (actually, it is set to
1091	 * ULLONG_MAX by default) - truncate upper end of the range
1092	 * carefully so as not to overflow.
1093	 */
1094	if (max_blocks - start_block < len)
1095		end_block = max_blocks - 1;
1096	else
1097		end_block = start_block + len - 1;
1098
1099	segnum = nilfs_get_segnum_of_block(nilfs, start_block);
1100	segnum_end = nilfs_get_segnum_of_block(nilfs, end_block);
1101
1102	down_read(&NILFS_MDT(sufile)->mi_sem);
1103
1104	while (segnum <= segnum_end) {
1105		n = nilfs_sufile_segment_usages_in_block(sufile, segnum,
1106				segnum_end);
1107
1108		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
1109							   &su_bh);
1110		if (ret < 0) {
1111			if (ret != -ENOENT)
1112				goto out_sem;
1113			/* hole */
1114			segnum += n;
1115			continue;
1116		}
1117
1118		kaddr = kmap_local_page(su_bh->b_page);
1119		su = nilfs_sufile_block_get_segment_usage(sufile, segnum,
1120				su_bh, kaddr);
1121		for (i = 0; i < n; ++i, ++segnum, su = (void *)su + susz) {
1122			if (!nilfs_segment_usage_clean(su))
1123				continue;
1124
1125			nilfs_get_segment_range(nilfs, segnum, &seg_start,
1126						&seg_end);
1127
1128			if (!nblocks) {
1129				/* start new extent */
1130				start = seg_start;
1131				nblocks = seg_end - seg_start + 1;
1132				continue;
1133			}
1134
1135			if (start + nblocks == seg_start) {
1136				/* add to previous extent */
1137				nblocks += seg_end - seg_start + 1;
1138				continue;
1139			}
1140
1141			/* discard previous extent */
1142			if (start < start_block) {
1143				nblocks -= start_block - start;
1144				start = start_block;
1145			}
1146
1147			if (nblocks >= minlen) {
1148				kunmap_local(kaddr);
1149
1150				ret = blkdev_issue_discard(nilfs->ns_bdev,
1151						start * sects_per_block,
1152						nblocks * sects_per_block,
1153						GFP_NOFS);
1154				if (ret < 0) {
1155					put_bh(su_bh);
1156					goto out_sem;
1157				}
1158
1159				ndiscarded += nblocks;
1160				kaddr = kmap_local_page(su_bh->b_page);
1161				su = nilfs_sufile_block_get_segment_usage(
1162					sufile, segnum, su_bh, kaddr);
1163			}
1164
1165			/* start new extent */
1166			start = seg_start;
1167			nblocks = seg_end - seg_start + 1;
1168		}
1169		kunmap_local(kaddr);
1170		put_bh(su_bh);
1171	}
1172
1173
1174	if (nblocks) {
1175		/* discard last extent */
1176		if (start < start_block) {
1177			nblocks -= start_block - start;
1178			start = start_block;
1179		}
1180		if (start + nblocks > end_block + 1)
1181			nblocks = end_block - start + 1;
1182
1183		if (nblocks >= minlen) {
1184			ret = blkdev_issue_discard(nilfs->ns_bdev,
1185					start * sects_per_block,
1186					nblocks * sects_per_block,
1187					GFP_NOFS);
1188			if (!ret)
1189				ndiscarded += nblocks;
1190		}
1191	}
1192
1193out_sem:
1194	up_read(&NILFS_MDT(sufile)->mi_sem);
1195
1196	range->len = ndiscarded << nilfs->ns_blocksize_bits;
1197	return ret;
1198}
1199
1200/**
1201 * nilfs_sufile_read - read or get sufile inode
1202 * @sb: super block instance
1203 * @susize: size of a segment usage entry
1204 * @raw_inode: on-disk sufile inode
1205 * @inodep: buffer to store the inode
1206 */
1207int nilfs_sufile_read(struct super_block *sb, size_t susize,
1208		      struct nilfs_inode *raw_inode, struct inode **inodep)
1209{
1210	struct inode *sufile;
1211	struct nilfs_sufile_info *sui;
1212	struct buffer_head *header_bh;
1213	struct nilfs_sufile_header *header;
1214	void *kaddr;
1215	int err;
1216
1217	if (susize > sb->s_blocksize) {
1218		nilfs_err(sb, "too large segment usage size: %zu bytes",
1219			  susize);
1220		return -EINVAL;
1221	} else if (susize < NILFS_MIN_SEGMENT_USAGE_SIZE) {
1222		nilfs_err(sb, "too small segment usage size: %zu bytes",
1223			  susize);
1224		return -EINVAL;
1225	}
1226
1227	sufile = nilfs_iget_locked(sb, NULL, NILFS_SUFILE_INO);
1228	if (unlikely(!sufile))
1229		return -ENOMEM;
1230	if (!(sufile->i_state & I_NEW))
1231		goto out;
1232
1233	err = nilfs_mdt_init(sufile, NILFS_MDT_GFP, sizeof(*sui));
1234	if (err)
1235		goto failed;
1236
1237	nilfs_mdt_set_entry_size(sufile, susize,
1238				 sizeof(struct nilfs_sufile_header));
1239
1240	err = nilfs_read_inode_common(sufile, raw_inode);
1241	if (err)
1242		goto failed;
1243
1244	err = nilfs_sufile_get_header_block(sufile, &header_bh);
1245	if (err)
1246		goto failed;
1247
1248	sui = NILFS_SUI(sufile);
1249	kaddr = kmap_local_page(header_bh->b_page);
1250	header = kaddr + bh_offset(header_bh);
1251	sui->ncleansegs = le64_to_cpu(header->sh_ncleansegs);
1252	kunmap_local(kaddr);
1253	brelse(header_bh);
1254
1255	sui->allocmax = nilfs_sufile_get_nsegments(sufile) - 1;
1256	sui->allocmin = 0;
1257
1258	unlock_new_inode(sufile);
1259 out:
1260	*inodep = sufile;
1261	return 0;
1262 failed:
1263	iget_failed(sufile);
1264	return err;
1265}
1266