xref: /linux/fs/nilfs2/sufile.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * sufile.c - NILFS segment usage file.
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  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Koji Sato <koji@osrg.net>.
21  * Revised by Ryusuke Konishi <ryusuke@osrg.net>.
22  */
23 
24 #include <linux/kernel.h>
25 #include <linux/fs.h>
26 #include <linux/string.h>
27 #include <linux/buffer_head.h>
28 #include <linux/errno.h>
29 #include <linux/nilfs2_fs.h>
30 #include "mdt.h"
31 #include "sufile.h"
32 
33 /**
34  * struct nilfs_sufile_info - on-memory private data of sufile
35  * @mi: on-memory private data of metadata file
36  * @ncleansegs: number of clean segments
37  * @allocmin: lower limit of allocatable segment range
38  * @allocmax: upper limit of allocatable segment range
39  */
40 struct nilfs_sufile_info {
41 	struct nilfs_mdt_info mi;
42 	unsigned long ncleansegs;/* number of clean segments */
43 	__u64 allocmin;		/* lower limit of allocatable segment range */
44 	__u64 allocmax;		/* upper limit of allocatable segment range */
45 };
46 
47 static inline struct nilfs_sufile_info *NILFS_SUI(struct inode *sufile)
48 {
49 	return (struct nilfs_sufile_info *)NILFS_MDT(sufile);
50 }
51 
52 static inline unsigned long
53 nilfs_sufile_segment_usages_per_block(const struct inode *sufile)
54 {
55 	return NILFS_MDT(sufile)->mi_entries_per_block;
56 }
57 
58 static unsigned long
59 nilfs_sufile_get_blkoff(const struct inode *sufile, __u64 segnum)
60 {
61 	__u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
62 	do_div(t, nilfs_sufile_segment_usages_per_block(sufile));
63 	return (unsigned long)t;
64 }
65 
66 static unsigned long
67 nilfs_sufile_get_offset(const struct inode *sufile, __u64 segnum)
68 {
69 	__u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
70 	return do_div(t, nilfs_sufile_segment_usages_per_block(sufile));
71 }
72 
73 static unsigned long
74 nilfs_sufile_segment_usages_in_block(const struct inode *sufile, __u64 curr,
75 				     __u64 max)
76 {
77 	return min_t(unsigned long,
78 		     nilfs_sufile_segment_usages_per_block(sufile) -
79 		     nilfs_sufile_get_offset(sufile, curr),
80 		     max - curr + 1);
81 }
82 
83 static struct nilfs_segment_usage *
84 nilfs_sufile_block_get_segment_usage(const struct inode *sufile, __u64 segnum,
85 				     struct buffer_head *bh, void *kaddr)
86 {
87 	return kaddr + bh_offset(bh) +
88 		nilfs_sufile_get_offset(sufile, segnum) *
89 		NILFS_MDT(sufile)->mi_entry_size;
90 }
91 
92 static inline int nilfs_sufile_get_header_block(struct inode *sufile,
93 						struct buffer_head **bhp)
94 {
95 	return nilfs_mdt_get_block(sufile, 0, 0, NULL, bhp);
96 }
97 
98 static inline int
99 nilfs_sufile_get_segment_usage_block(struct inode *sufile, __u64 segnum,
100 				     int create, struct buffer_head **bhp)
101 {
102 	return nilfs_mdt_get_block(sufile,
103 				   nilfs_sufile_get_blkoff(sufile, segnum),
104 				   create, NULL, bhp);
105 }
106 
107 static int nilfs_sufile_delete_segment_usage_block(struct inode *sufile,
108 						   __u64 segnum)
109 {
110 	return nilfs_mdt_delete_block(sufile,
111 				      nilfs_sufile_get_blkoff(sufile, segnum));
112 }
113 
114 static void nilfs_sufile_mod_counter(struct buffer_head *header_bh,
115 				     u64 ncleanadd, u64 ndirtyadd)
116 {
117 	struct nilfs_sufile_header *header;
118 	void *kaddr;
119 
120 	kaddr = kmap_atomic(header_bh->b_page);
121 	header = kaddr + bh_offset(header_bh);
122 	le64_add_cpu(&header->sh_ncleansegs, ncleanadd);
123 	le64_add_cpu(&header->sh_ndirtysegs, ndirtyadd);
124 	kunmap_atomic(kaddr);
125 
126 	mark_buffer_dirty(header_bh);
127 }
128 
129 /**
130  * nilfs_sufile_get_ncleansegs - return the number of clean segments
131  * @sufile: inode of segment usage file
132  */
133 unsigned long nilfs_sufile_get_ncleansegs(struct inode *sufile)
134 {
135 	return NILFS_SUI(sufile)->ncleansegs;
136 }
137 
138 /**
139  * nilfs_sufile_updatev - modify multiple segment usages at a time
140  * @sufile: inode of segment usage file
141  * @segnumv: array of segment numbers
142  * @nsegs: size of @segnumv array
143  * @create: creation flag
144  * @ndone: place to store number of modified segments on @segnumv
145  * @dofunc: primitive operation for the update
146  *
147  * Description: nilfs_sufile_updatev() repeatedly calls @dofunc
148  * against the given array of segments.  The @dofunc is called with
149  * buffers of a header block and the sufile block in which the target
150  * segment usage entry is contained.  If @ndone is given, the number
151  * of successfully modified segments from the head is stored in the
152  * place @ndone points to.
153  *
154  * Return Value: On success, zero is returned.  On error, one of the
155  * following negative error codes is returned.
156  *
157  * %-EIO - I/O error.
158  *
159  * %-ENOMEM - Insufficient amount of memory available.
160  *
161  * %-ENOENT - Given segment usage is in hole block (may be returned if
162  *            @create is zero)
163  *
164  * %-EINVAL - Invalid segment usage number
165  */
166 int nilfs_sufile_updatev(struct inode *sufile, __u64 *segnumv, size_t nsegs,
167 			 int create, size_t *ndone,
168 			 void (*dofunc)(struct inode *, __u64,
169 					struct buffer_head *,
170 					struct buffer_head *))
171 {
172 	struct buffer_head *header_bh, *bh;
173 	unsigned long blkoff, prev_blkoff;
174 	__u64 *seg;
175 	size_t nerr = 0, n = 0;
176 	int ret = 0;
177 
178 	if (unlikely(nsegs == 0))
179 		goto out;
180 
181 	down_write(&NILFS_MDT(sufile)->mi_sem);
182 	for (seg = segnumv; seg < segnumv + nsegs; seg++) {
183 		if (unlikely(*seg >= nilfs_sufile_get_nsegments(sufile))) {
184 			printk(KERN_WARNING
185 			       "%s: invalid segment number: %llu\n", __func__,
186 			       (unsigned long long)*seg);
187 			nerr++;
188 		}
189 	}
190 	if (nerr > 0) {
191 		ret = -EINVAL;
192 		goto out_sem;
193 	}
194 
195 	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
196 	if (ret < 0)
197 		goto out_sem;
198 
199 	seg = segnumv;
200 	blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
201 	ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
202 	if (ret < 0)
203 		goto out_header;
204 
205 	for (;;) {
206 		dofunc(sufile, *seg, header_bh, bh);
207 
208 		if (++seg >= segnumv + nsegs)
209 			break;
210 		prev_blkoff = blkoff;
211 		blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
212 		if (blkoff == prev_blkoff)
213 			continue;
214 
215 		/* get different block */
216 		brelse(bh);
217 		ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
218 		if (unlikely(ret < 0))
219 			goto out_header;
220 	}
221 	brelse(bh);
222 
223  out_header:
224 	n = seg - segnumv;
225 	brelse(header_bh);
226  out_sem:
227 	up_write(&NILFS_MDT(sufile)->mi_sem);
228  out:
229 	if (ndone)
230 		*ndone = n;
231 	return ret;
232 }
233 
234 int nilfs_sufile_update(struct inode *sufile, __u64 segnum, int create,
235 			void (*dofunc)(struct inode *, __u64,
236 				       struct buffer_head *,
237 				       struct buffer_head *))
238 {
239 	struct buffer_head *header_bh, *bh;
240 	int ret;
241 
242 	if (unlikely(segnum >= nilfs_sufile_get_nsegments(sufile))) {
243 		printk(KERN_WARNING "%s: invalid segment number: %llu\n",
244 		       __func__, (unsigned long long)segnum);
245 		return -EINVAL;
246 	}
247 	down_write(&NILFS_MDT(sufile)->mi_sem);
248 
249 	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
250 	if (ret < 0)
251 		goto out_sem;
252 
253 	ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, create, &bh);
254 	if (!ret) {
255 		dofunc(sufile, segnum, header_bh, bh);
256 		brelse(bh);
257 	}
258 	brelse(header_bh);
259 
260  out_sem:
261 	up_write(&NILFS_MDT(sufile)->mi_sem);
262 	return ret;
263 }
264 
265 /**
266  * nilfs_sufile_set_alloc_range - limit range of segment to be allocated
267  * @sufile: inode of segment usage file
268  * @start: minimum segment number of allocatable region (inclusive)
269  * @end: maximum segment number of allocatable region (inclusive)
270  *
271  * Return Value: On success, 0 is returned.  On error, one of the
272  * following negative error codes is returned.
273  *
274  * %-ERANGE - invalid segment region
275  */
276 int nilfs_sufile_set_alloc_range(struct inode *sufile, __u64 start, __u64 end)
277 {
278 	struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
279 	__u64 nsegs;
280 	int ret = -ERANGE;
281 
282 	down_write(&NILFS_MDT(sufile)->mi_sem);
283 	nsegs = nilfs_sufile_get_nsegments(sufile);
284 
285 	if (start <= end && end < nsegs) {
286 		sui->allocmin = start;
287 		sui->allocmax = end;
288 		ret = 0;
289 	}
290 	up_write(&NILFS_MDT(sufile)->mi_sem);
291 	return ret;
292 }
293 
294 /**
295  * nilfs_sufile_alloc - allocate a segment
296  * @sufile: inode of segment usage file
297  * @segnump: pointer to segment number
298  *
299  * Description: nilfs_sufile_alloc() allocates a clean segment.
300  *
301  * Return Value: On success, 0 is returned and the segment number of the
302  * allocated segment is stored in the place pointed by @segnump. On error, one
303  * of the following negative error codes is returned.
304  *
305  * %-EIO - I/O error.
306  *
307  * %-ENOMEM - Insufficient amount of memory available.
308  *
309  * %-ENOSPC - No clean segment left.
310  */
311 int nilfs_sufile_alloc(struct inode *sufile, __u64 *segnump)
312 {
313 	struct buffer_head *header_bh, *su_bh;
314 	struct nilfs_sufile_header *header;
315 	struct nilfs_segment_usage *su;
316 	struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
317 	size_t susz = NILFS_MDT(sufile)->mi_entry_size;
318 	__u64 segnum, maxsegnum, last_alloc;
319 	void *kaddr;
320 	unsigned long nsegments, ncleansegs, nsus, cnt;
321 	int ret, j;
322 
323 	down_write(&NILFS_MDT(sufile)->mi_sem);
324 
325 	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
326 	if (ret < 0)
327 		goto out_sem;
328 	kaddr = kmap_atomic(header_bh->b_page);
329 	header = kaddr + bh_offset(header_bh);
330 	ncleansegs = le64_to_cpu(header->sh_ncleansegs);
331 	last_alloc = le64_to_cpu(header->sh_last_alloc);
332 	kunmap_atomic(kaddr);
333 
334 	nsegments = nilfs_sufile_get_nsegments(sufile);
335 	maxsegnum = sui->allocmax;
336 	segnum = last_alloc + 1;
337 	if (segnum < sui->allocmin || segnum > sui->allocmax)
338 		segnum = sui->allocmin;
339 
340 	for (cnt = 0; cnt < nsegments; cnt += nsus) {
341 		if (segnum > maxsegnum) {
342 			if (cnt < sui->allocmax - sui->allocmin + 1) {
343 				/*
344 				 * wrap around in the limited region.
345 				 * if allocation started from
346 				 * sui->allocmin, this never happens.
347 				 */
348 				segnum = sui->allocmin;
349 				maxsegnum = last_alloc;
350 			} else if (segnum > sui->allocmin &&
351 				   sui->allocmax + 1 < nsegments) {
352 				segnum = sui->allocmax + 1;
353 				maxsegnum = nsegments - 1;
354 			} else if (sui->allocmin > 0)  {
355 				segnum = 0;
356 				maxsegnum = sui->allocmin - 1;
357 			} else {
358 				break; /* never happens */
359 			}
360 		}
361 		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 1,
362 							   &su_bh);
363 		if (ret < 0)
364 			goto out_header;
365 		kaddr = kmap_atomic(su_bh->b_page);
366 		su = nilfs_sufile_block_get_segment_usage(
367 			sufile, segnum, su_bh, kaddr);
368 
369 		nsus = nilfs_sufile_segment_usages_in_block(
370 			sufile, segnum, maxsegnum);
371 		for (j = 0; j < nsus; j++, su = (void *)su + susz, segnum++) {
372 			if (!nilfs_segment_usage_clean(su))
373 				continue;
374 			/* found a clean segment */
375 			nilfs_segment_usage_set_dirty(su);
376 			kunmap_atomic(kaddr);
377 
378 			kaddr = kmap_atomic(header_bh->b_page);
379 			header = kaddr + bh_offset(header_bh);
380 			le64_add_cpu(&header->sh_ncleansegs, -1);
381 			le64_add_cpu(&header->sh_ndirtysegs, 1);
382 			header->sh_last_alloc = cpu_to_le64(segnum);
383 			kunmap_atomic(kaddr);
384 
385 			sui->ncleansegs--;
386 			mark_buffer_dirty(header_bh);
387 			mark_buffer_dirty(su_bh);
388 			nilfs_mdt_mark_dirty(sufile);
389 			brelse(su_bh);
390 			*segnump = segnum;
391 			goto out_header;
392 		}
393 
394 		kunmap_atomic(kaddr);
395 		brelse(su_bh);
396 	}
397 
398 	/* no segments left */
399 	ret = -ENOSPC;
400 
401  out_header:
402 	brelse(header_bh);
403 
404  out_sem:
405 	up_write(&NILFS_MDT(sufile)->mi_sem);
406 	return ret;
407 }
408 
409 void nilfs_sufile_do_cancel_free(struct inode *sufile, __u64 segnum,
410 				 struct buffer_head *header_bh,
411 				 struct buffer_head *su_bh)
412 {
413 	struct nilfs_segment_usage *su;
414 	void *kaddr;
415 
416 	kaddr = kmap_atomic(su_bh->b_page);
417 	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
418 	if (unlikely(!nilfs_segment_usage_clean(su))) {
419 		printk(KERN_WARNING "%s: segment %llu must be clean\n",
420 		       __func__, (unsigned long long)segnum);
421 		kunmap_atomic(kaddr);
422 		return;
423 	}
424 	nilfs_segment_usage_set_dirty(su);
425 	kunmap_atomic(kaddr);
426 
427 	nilfs_sufile_mod_counter(header_bh, -1, 1);
428 	NILFS_SUI(sufile)->ncleansegs--;
429 
430 	mark_buffer_dirty(su_bh);
431 	nilfs_mdt_mark_dirty(sufile);
432 }
433 
434 void nilfs_sufile_do_scrap(struct inode *sufile, __u64 segnum,
435 			   struct buffer_head *header_bh,
436 			   struct buffer_head *su_bh)
437 {
438 	struct nilfs_segment_usage *su;
439 	void *kaddr;
440 	int clean, dirty;
441 
442 	kaddr = kmap_atomic(su_bh->b_page);
443 	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
444 	if (su->su_flags == cpu_to_le32(1UL << NILFS_SEGMENT_USAGE_DIRTY) &&
445 	    su->su_nblocks == cpu_to_le32(0)) {
446 		kunmap_atomic(kaddr);
447 		return;
448 	}
449 	clean = nilfs_segment_usage_clean(su);
450 	dirty = nilfs_segment_usage_dirty(su);
451 
452 	/* make the segment garbage */
453 	su->su_lastmod = cpu_to_le64(0);
454 	su->su_nblocks = cpu_to_le32(0);
455 	su->su_flags = cpu_to_le32(1UL << NILFS_SEGMENT_USAGE_DIRTY);
456 	kunmap_atomic(kaddr);
457 
458 	nilfs_sufile_mod_counter(header_bh, clean ? (u64)-1 : 0, dirty ? 0 : 1);
459 	NILFS_SUI(sufile)->ncleansegs -= clean;
460 
461 	mark_buffer_dirty(su_bh);
462 	nilfs_mdt_mark_dirty(sufile);
463 }
464 
465 void nilfs_sufile_do_free(struct inode *sufile, __u64 segnum,
466 			  struct buffer_head *header_bh,
467 			  struct buffer_head *su_bh)
468 {
469 	struct nilfs_segment_usage *su;
470 	void *kaddr;
471 	int sudirty;
472 
473 	kaddr = kmap_atomic(su_bh->b_page);
474 	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
475 	if (nilfs_segment_usage_clean(su)) {
476 		printk(KERN_WARNING "%s: segment %llu is already clean\n",
477 		       __func__, (unsigned long long)segnum);
478 		kunmap_atomic(kaddr);
479 		return;
480 	}
481 	WARN_ON(nilfs_segment_usage_error(su));
482 	WARN_ON(!nilfs_segment_usage_dirty(su));
483 
484 	sudirty = nilfs_segment_usage_dirty(su);
485 	nilfs_segment_usage_set_clean(su);
486 	kunmap_atomic(kaddr);
487 	mark_buffer_dirty(su_bh);
488 
489 	nilfs_sufile_mod_counter(header_bh, 1, sudirty ? (u64)-1 : 0);
490 	NILFS_SUI(sufile)->ncleansegs++;
491 
492 	nilfs_mdt_mark_dirty(sufile);
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  */
500 int nilfs_sufile_mark_dirty(struct inode *sufile, __u64 segnum)
501 {
502 	struct buffer_head *bh;
503 	int ret;
504 
505 	ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
506 	if (!ret) {
507 		mark_buffer_dirty(bh);
508 		nilfs_mdt_mark_dirty(sufile);
509 		brelse(bh);
510 	}
511 	return ret;
512 }
513 
514 /**
515  * nilfs_sufile_set_segment_usage - set usage of a segment
516  * @sufile: inode of segment usage file
517  * @segnum: segment number
518  * @nblocks: number of live blocks in the segment
519  * @modtime: modification time (option)
520  */
521 int nilfs_sufile_set_segment_usage(struct inode *sufile, __u64 segnum,
522 				   unsigned long nblocks, time_t modtime)
523 {
524 	struct buffer_head *bh;
525 	struct nilfs_segment_usage *su;
526 	void *kaddr;
527 	int ret;
528 
529 	down_write(&NILFS_MDT(sufile)->mi_sem);
530 	ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
531 	if (ret < 0)
532 		goto out_sem;
533 
534 	kaddr = kmap_atomic(bh->b_page);
535 	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);
536 	WARN_ON(nilfs_segment_usage_error(su));
537 	if (modtime)
538 		su->su_lastmod = cpu_to_le64(modtime);
539 	su->su_nblocks = cpu_to_le32(nblocks);
540 	kunmap_atomic(kaddr);
541 
542 	mark_buffer_dirty(bh);
543 	nilfs_mdt_mark_dirty(sufile);
544 	brelse(bh);
545 
546  out_sem:
547 	up_write(&NILFS_MDT(sufile)->mi_sem);
548 	return ret;
549 }
550 
551 /**
552  * nilfs_sufile_get_stat - get segment usage statistics
553  * @sufile: inode of segment usage file
554  * @stat: pointer to a structure of segment usage statistics
555  *
556  * Description: nilfs_sufile_get_stat() returns information about segment
557  * usage.
558  *
559  * Return Value: On success, 0 is returned, and segment usage information is
560  * stored in the place pointed by @stat. On error, one of the following
561  * negative error codes is returned.
562  *
563  * %-EIO - I/O error.
564  *
565  * %-ENOMEM - Insufficient amount of memory available.
566  */
567 int nilfs_sufile_get_stat(struct inode *sufile, struct nilfs_sustat *sustat)
568 {
569 	struct buffer_head *header_bh;
570 	struct nilfs_sufile_header *header;
571 	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
572 	void *kaddr;
573 	int ret;
574 
575 	down_read(&NILFS_MDT(sufile)->mi_sem);
576 
577 	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
578 	if (ret < 0)
579 		goto out_sem;
580 
581 	kaddr = kmap_atomic(header_bh->b_page);
582 	header = kaddr + bh_offset(header_bh);
583 	sustat->ss_nsegs = nilfs_sufile_get_nsegments(sufile);
584 	sustat->ss_ncleansegs = le64_to_cpu(header->sh_ncleansegs);
585 	sustat->ss_ndirtysegs = le64_to_cpu(header->sh_ndirtysegs);
586 	sustat->ss_ctime = nilfs->ns_ctime;
587 	sustat->ss_nongc_ctime = nilfs->ns_nongc_ctime;
588 	spin_lock(&nilfs->ns_last_segment_lock);
589 	sustat->ss_prot_seq = nilfs->ns_prot_seq;
590 	spin_unlock(&nilfs->ns_last_segment_lock);
591 	kunmap_atomic(kaddr);
592 	brelse(header_bh);
593 
594  out_sem:
595 	up_read(&NILFS_MDT(sufile)->mi_sem);
596 	return ret;
597 }
598 
599 void nilfs_sufile_do_set_error(struct inode *sufile, __u64 segnum,
600 			       struct buffer_head *header_bh,
601 			       struct buffer_head *su_bh)
602 {
603 	struct nilfs_segment_usage *su;
604 	void *kaddr;
605 	int suclean;
606 
607 	kaddr = kmap_atomic(su_bh->b_page);
608 	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
609 	if (nilfs_segment_usage_error(su)) {
610 		kunmap_atomic(kaddr);
611 		return;
612 	}
613 	suclean = nilfs_segment_usage_clean(su);
614 	nilfs_segment_usage_set_error(su);
615 	kunmap_atomic(kaddr);
616 
617 	if (suclean) {
618 		nilfs_sufile_mod_counter(header_bh, -1, 0);
619 		NILFS_SUI(sufile)->ncleansegs--;
620 	}
621 	mark_buffer_dirty(su_bh);
622 	nilfs_mdt_mark_dirty(sufile);
623 }
624 
625 /**
626   * nilfs_sufile_truncate_range - truncate range of segment array
627   * @sufile: inode of segment usage file
628   * @start: start segment number (inclusive)
629   * @end: end segment number (inclusive)
630   *
631   * Return Value: On success, 0 is returned.  On error, one of the
632   * following negative error codes is returned.
633   *
634   * %-EIO - I/O error.
635   *
636   * %-ENOMEM - Insufficient amount of memory available.
637   *
638   * %-EINVAL - Invalid number of segments specified
639   *
640   * %-EBUSY - Dirty or active segments are present in the range
641   */
642 static int nilfs_sufile_truncate_range(struct inode *sufile,
643 				       __u64 start, __u64 end)
644 {
645 	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
646 	struct buffer_head *header_bh;
647 	struct buffer_head *su_bh;
648 	struct nilfs_segment_usage *su, *su2;
649 	size_t susz = NILFS_MDT(sufile)->mi_entry_size;
650 	unsigned long segusages_per_block;
651 	unsigned long nsegs, ncleaned;
652 	__u64 segnum;
653 	void *kaddr;
654 	ssize_t n, nc;
655 	int ret;
656 	int j;
657 
658 	nsegs = nilfs_sufile_get_nsegments(sufile);
659 
660 	ret = -EINVAL;
661 	if (start > end || start >= nsegs)
662 		goto out;
663 
664 	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
665 	if (ret < 0)
666 		goto out;
667 
668 	segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
669 	ncleaned = 0;
670 
671 	for (segnum = start; segnum <= end; segnum += n) {
672 		n = min_t(unsigned long,
673 			  segusages_per_block -
674 				  nilfs_sufile_get_offset(sufile, segnum),
675 			  end - segnum + 1);
676 		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
677 							   &su_bh);
678 		if (ret < 0) {
679 			if (ret != -ENOENT)
680 				goto out_header;
681 			/* hole */
682 			continue;
683 		}
684 		kaddr = kmap_atomic(su_bh->b_page);
685 		su = nilfs_sufile_block_get_segment_usage(
686 			sufile, segnum, su_bh, kaddr);
687 		su2 = su;
688 		for (j = 0; j < n; j++, su = (void *)su + susz) {
689 			if ((le32_to_cpu(su->su_flags) &
690 			     ~(1UL << NILFS_SEGMENT_USAGE_ERROR)) ||
691 			    nilfs_segment_is_active(nilfs, segnum + j)) {
692 				ret = -EBUSY;
693 				kunmap_atomic(kaddr);
694 				brelse(su_bh);
695 				goto out_header;
696 			}
697 		}
698 		nc = 0;
699 		for (su = su2, j = 0; j < n; j++, su = (void *)su + susz) {
700 			if (nilfs_segment_usage_error(su)) {
701 				nilfs_segment_usage_set_clean(su);
702 				nc++;
703 			}
704 		}
705 		kunmap_atomic(kaddr);
706 		if (nc > 0) {
707 			mark_buffer_dirty(su_bh);
708 			ncleaned += nc;
709 		}
710 		brelse(su_bh);
711 
712 		if (n == segusages_per_block) {
713 			/* make hole */
714 			nilfs_sufile_delete_segment_usage_block(sufile, segnum);
715 		}
716 	}
717 	ret = 0;
718 
719 out_header:
720 	if (ncleaned > 0) {
721 		NILFS_SUI(sufile)->ncleansegs += ncleaned;
722 		nilfs_sufile_mod_counter(header_bh, ncleaned, 0);
723 		nilfs_mdt_mark_dirty(sufile);
724 	}
725 	brelse(header_bh);
726 out:
727 	return ret;
728 }
729 
730 /**
731  * nilfs_sufile_resize - resize segment array
732  * @sufile: inode of segment usage file
733  * @newnsegs: new number of segments
734  *
735  * Return Value: On success, 0 is returned.  On error, one of the
736  * following negative error codes is returned.
737  *
738  * %-EIO - I/O error.
739  *
740  * %-ENOMEM - Insufficient amount of memory available.
741  *
742  * %-ENOSPC - Enough free space is not left for shrinking
743  *
744  * %-EBUSY - Dirty or active segments exist in the region to be truncated
745  */
746 int nilfs_sufile_resize(struct inode *sufile, __u64 newnsegs)
747 {
748 	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
749 	struct buffer_head *header_bh;
750 	struct nilfs_sufile_header *header;
751 	struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
752 	void *kaddr;
753 	unsigned long nsegs, nrsvsegs;
754 	int ret = 0;
755 
756 	down_write(&NILFS_MDT(sufile)->mi_sem);
757 
758 	nsegs = nilfs_sufile_get_nsegments(sufile);
759 	if (nsegs == newnsegs)
760 		goto out;
761 
762 	ret = -ENOSPC;
763 	nrsvsegs = nilfs_nrsvsegs(nilfs, newnsegs);
764 	if (newnsegs < nsegs && nsegs - newnsegs + nrsvsegs > sui->ncleansegs)
765 		goto out;
766 
767 	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
768 	if (ret < 0)
769 		goto out;
770 
771 	if (newnsegs > nsegs) {
772 		sui->ncleansegs += newnsegs - nsegs;
773 	} else /* newnsegs < nsegs */ {
774 		ret = nilfs_sufile_truncate_range(sufile, newnsegs, nsegs - 1);
775 		if (ret < 0)
776 			goto out_header;
777 
778 		sui->ncleansegs -= nsegs - newnsegs;
779 	}
780 
781 	kaddr = kmap_atomic(header_bh->b_page);
782 	header = kaddr + bh_offset(header_bh);
783 	header->sh_ncleansegs = cpu_to_le64(sui->ncleansegs);
784 	kunmap_atomic(kaddr);
785 
786 	mark_buffer_dirty(header_bh);
787 	nilfs_mdt_mark_dirty(sufile);
788 	nilfs_set_nsegments(nilfs, newnsegs);
789 
790 out_header:
791 	brelse(header_bh);
792 out:
793 	up_write(&NILFS_MDT(sufile)->mi_sem);
794 	return ret;
795 }
796 
797 /**
798  * nilfs_sufile_get_suinfo -
799  * @sufile: inode of segment usage file
800  * @segnum: segment number to start looking
801  * @buf: array of suinfo
802  * @sisz: byte size of suinfo
803  * @nsi: size of suinfo array
804  *
805  * Description:
806  *
807  * Return Value: On success, 0 is returned and .... On error, one of the
808  * following negative error codes is returned.
809  *
810  * %-EIO - I/O error.
811  *
812  * %-ENOMEM - Insufficient amount of memory available.
813  */
814 ssize_t nilfs_sufile_get_suinfo(struct inode *sufile, __u64 segnum, void *buf,
815 				unsigned sisz, size_t nsi)
816 {
817 	struct buffer_head *su_bh;
818 	struct nilfs_segment_usage *su;
819 	struct nilfs_suinfo *si = buf;
820 	size_t susz = NILFS_MDT(sufile)->mi_entry_size;
821 	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
822 	void *kaddr;
823 	unsigned long nsegs, segusages_per_block;
824 	ssize_t n;
825 	int ret, i, j;
826 
827 	down_read(&NILFS_MDT(sufile)->mi_sem);
828 
829 	segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
830 	nsegs = min_t(unsigned long,
831 		      nilfs_sufile_get_nsegments(sufile) - segnum,
832 		      nsi);
833 	for (i = 0; i < nsegs; i += n, segnum += n) {
834 		n = min_t(unsigned long,
835 			  segusages_per_block -
836 				  nilfs_sufile_get_offset(sufile, segnum),
837 			  nsegs - i);
838 		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
839 							   &su_bh);
840 		if (ret < 0) {
841 			if (ret != -ENOENT)
842 				goto out;
843 			/* hole */
844 			memset(si, 0, sisz * n);
845 			si = (void *)si + sisz * n;
846 			continue;
847 		}
848 
849 		kaddr = kmap_atomic(su_bh->b_page);
850 		su = nilfs_sufile_block_get_segment_usage(
851 			sufile, segnum, su_bh, kaddr);
852 		for (j = 0; j < n;
853 		     j++, su = (void *)su + susz, si = (void *)si + sisz) {
854 			si->sui_lastmod = le64_to_cpu(su->su_lastmod);
855 			si->sui_nblocks = le32_to_cpu(su->su_nblocks);
856 			si->sui_flags = le32_to_cpu(su->su_flags) &
857 				~(1UL << NILFS_SEGMENT_USAGE_ACTIVE);
858 			if (nilfs_segment_is_active(nilfs, segnum + j))
859 				si->sui_flags |=
860 					(1UL << NILFS_SEGMENT_USAGE_ACTIVE);
861 		}
862 		kunmap_atomic(kaddr);
863 		brelse(su_bh);
864 	}
865 	ret = nsegs;
866 
867  out:
868 	up_read(&NILFS_MDT(sufile)->mi_sem);
869 	return ret;
870 }
871 
872 /**
873  * nilfs_sufile_set_suinfo - sets segment usage info
874  * @sufile: inode of segment usage file
875  * @buf: array of suinfo_update
876  * @supsz: byte size of suinfo_update
877  * @nsup: size of suinfo_update array
878  *
879  * Description: Takes an array of nilfs_suinfo_update structs and updates
880  * segment usage accordingly. Only the fields indicated by the sup_flags
881  * are updated.
882  *
883  * Return Value: On success, 0 is returned. On error, one of the
884  * following negative error codes is returned.
885  *
886  * %-EIO - I/O error.
887  *
888  * %-ENOMEM - Insufficient amount of memory available.
889  *
890  * %-EINVAL - Invalid values in input (segment number, flags or nblocks)
891  */
892 ssize_t nilfs_sufile_set_suinfo(struct inode *sufile, void *buf,
893 				unsigned supsz, size_t nsup)
894 {
895 	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
896 	struct buffer_head *header_bh, *bh;
897 	struct nilfs_suinfo_update *sup, *supend = buf + supsz * nsup;
898 	struct nilfs_segment_usage *su;
899 	void *kaddr;
900 	unsigned long blkoff, prev_blkoff;
901 	int cleansi, cleansu, dirtysi, dirtysu;
902 	long ncleaned = 0, ndirtied = 0;
903 	int ret = 0;
904 
905 	if (unlikely(nsup == 0))
906 		return ret;
907 
908 	for (sup = buf; sup < supend; sup = (void *)sup + supsz) {
909 		if (sup->sup_segnum >= nilfs->ns_nsegments
910 			|| (sup->sup_flags &
911 				(~0UL << __NR_NILFS_SUINFO_UPDATE_FIELDS))
912 			|| (nilfs_suinfo_update_nblocks(sup) &&
913 				sup->sup_sui.sui_nblocks >
914 				nilfs->ns_blocks_per_segment))
915 			return -EINVAL;
916 	}
917 
918 	down_write(&NILFS_MDT(sufile)->mi_sem);
919 
920 	ret = nilfs_sufile_get_header_block(sufile, &header_bh);
921 	if (ret < 0)
922 		goto out_sem;
923 
924 	sup = buf;
925 	blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
926 	ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
927 	if (ret < 0)
928 		goto out_header;
929 
930 	for (;;) {
931 		kaddr = kmap_atomic(bh->b_page);
932 		su = nilfs_sufile_block_get_segment_usage(
933 			sufile, sup->sup_segnum, bh, kaddr);
934 
935 		if (nilfs_suinfo_update_lastmod(sup))
936 			su->su_lastmod = cpu_to_le64(sup->sup_sui.sui_lastmod);
937 
938 		if (nilfs_suinfo_update_nblocks(sup))
939 			su->su_nblocks = cpu_to_le32(sup->sup_sui.sui_nblocks);
940 
941 		if (nilfs_suinfo_update_flags(sup)) {
942 			/*
943 			 * Active flag is a virtual flag projected by running
944 			 * nilfs kernel code - drop it not to write it to
945 			 * disk.
946 			 */
947 			sup->sup_sui.sui_flags &=
948 					~(1UL << NILFS_SEGMENT_USAGE_ACTIVE);
949 
950 			cleansi = nilfs_suinfo_clean(&sup->sup_sui);
951 			cleansu = nilfs_segment_usage_clean(su);
952 			dirtysi = nilfs_suinfo_dirty(&sup->sup_sui);
953 			dirtysu = nilfs_segment_usage_dirty(su);
954 
955 			if (cleansi && !cleansu)
956 				++ncleaned;
957 			else if (!cleansi && cleansu)
958 				--ncleaned;
959 
960 			if (dirtysi && !dirtysu)
961 				++ndirtied;
962 			else if (!dirtysi && dirtysu)
963 				--ndirtied;
964 
965 			su->su_flags = cpu_to_le32(sup->sup_sui.sui_flags);
966 		}
967 
968 		kunmap_atomic(kaddr);
969 
970 		sup = (void *)sup + supsz;
971 		if (sup >= supend)
972 			break;
973 
974 		prev_blkoff = blkoff;
975 		blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
976 		if (blkoff == prev_blkoff)
977 			continue;
978 
979 		/* get different block */
980 		mark_buffer_dirty(bh);
981 		put_bh(bh);
982 		ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
983 		if (unlikely(ret < 0))
984 			goto out_mark;
985 	}
986 	mark_buffer_dirty(bh);
987 	put_bh(bh);
988 
989  out_mark:
990 	if (ncleaned || ndirtied) {
991 		nilfs_sufile_mod_counter(header_bh, (u64)ncleaned,
992 				(u64)ndirtied);
993 		NILFS_SUI(sufile)->ncleansegs += ncleaned;
994 	}
995 	nilfs_mdt_mark_dirty(sufile);
996  out_header:
997 	put_bh(header_bh);
998  out_sem:
999 	up_write(&NILFS_MDT(sufile)->mi_sem);
1000 	return ret;
1001 }
1002 
1003 /**
1004  * nilfs_sufile_trim_fs() - trim ioctl handle function
1005  * @sufile: inode of segment usage file
1006  * @range: fstrim_range structure
1007  *
1008  * start:	First Byte to trim
1009  * len:		number of Bytes to trim from start
1010  * minlen:	minimum extent length in Bytes
1011  *
1012  * Decription: nilfs_sufile_trim_fs goes through all segments containing bytes
1013  * from start to start+len. start is rounded up to the next block boundary
1014  * and start+len is rounded down. For each clean segment blkdev_issue_discard
1015  * function is invoked.
1016  *
1017  * Return Value: On success, 0 is returned or negative error code, otherwise.
1018  */
1019 int nilfs_sufile_trim_fs(struct inode *sufile, struct fstrim_range *range)
1020 {
1021 	struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
1022 	struct buffer_head *su_bh;
1023 	struct nilfs_segment_usage *su;
1024 	void *kaddr;
1025 	size_t n, i, susz = NILFS_MDT(sufile)->mi_entry_size;
1026 	sector_t seg_start, seg_end, start_block, end_block;
1027 	sector_t start = 0, nblocks = 0;
1028 	u64 segnum, segnum_end, minlen, len, max_blocks, ndiscarded = 0;
1029 	int ret = 0;
1030 	unsigned int sects_per_block;
1031 
1032 	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
1033 			bdev_logical_block_size(nilfs->ns_bdev);
1034 	len = range->len >> nilfs->ns_blocksize_bits;
1035 	minlen = range->minlen >> nilfs->ns_blocksize_bits;
1036 	max_blocks = ((u64)nilfs->ns_nsegments * nilfs->ns_blocks_per_segment);
1037 
1038 	if (!len || range->start >= max_blocks << nilfs->ns_blocksize_bits)
1039 		return -EINVAL;
1040 
1041 	start_block = (range->start + nilfs->ns_blocksize - 1) >>
1042 			nilfs->ns_blocksize_bits;
1043 
1044 	/*
1045 	 * range->len can be very large (actually, it is set to
1046 	 * ULLONG_MAX by default) - truncate upper end of the range
1047 	 * carefully so as not to overflow.
1048 	 */
1049 	if (max_blocks - start_block < len)
1050 		end_block = max_blocks - 1;
1051 	else
1052 		end_block = start_block + len - 1;
1053 
1054 	segnum = nilfs_get_segnum_of_block(nilfs, start_block);
1055 	segnum_end = nilfs_get_segnum_of_block(nilfs, end_block);
1056 
1057 	down_read(&NILFS_MDT(sufile)->mi_sem);
1058 
1059 	while (segnum <= segnum_end) {
1060 		n = nilfs_sufile_segment_usages_in_block(sufile, segnum,
1061 				segnum_end);
1062 
1063 		ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
1064 							   &su_bh);
1065 		if (ret < 0) {
1066 			if (ret != -ENOENT)
1067 				goto out_sem;
1068 			/* hole */
1069 			segnum += n;
1070 			continue;
1071 		}
1072 
1073 		kaddr = kmap_atomic(su_bh->b_page);
1074 		su = nilfs_sufile_block_get_segment_usage(sufile, segnum,
1075 				su_bh, kaddr);
1076 		for (i = 0; i < n; ++i, ++segnum, su = (void *)su + susz) {
1077 			if (!nilfs_segment_usage_clean(su))
1078 				continue;
1079 
1080 			nilfs_get_segment_range(nilfs, segnum, &seg_start,
1081 						&seg_end);
1082 
1083 			if (!nblocks) {
1084 				/* start new extent */
1085 				start = seg_start;
1086 				nblocks = seg_end - seg_start + 1;
1087 				continue;
1088 			}
1089 
1090 			if (start + nblocks == seg_start) {
1091 				/* add to previous extent */
1092 				nblocks += seg_end - seg_start + 1;
1093 				continue;
1094 			}
1095 
1096 			/* discard previous extent */
1097 			if (start < start_block) {
1098 				nblocks -= start_block - start;
1099 				start = start_block;
1100 			}
1101 
1102 			if (nblocks >= minlen) {
1103 				kunmap_atomic(kaddr);
1104 
1105 				ret = blkdev_issue_discard(nilfs->ns_bdev,
1106 						start * sects_per_block,
1107 						nblocks * sects_per_block,
1108 						GFP_NOFS, 0);
1109 				if (ret < 0) {
1110 					put_bh(su_bh);
1111 					goto out_sem;
1112 				}
1113 
1114 				ndiscarded += nblocks;
1115 				kaddr = kmap_atomic(su_bh->b_page);
1116 				su = nilfs_sufile_block_get_segment_usage(
1117 					sufile, segnum, su_bh, kaddr);
1118 			}
1119 
1120 			/* start new extent */
1121 			start = seg_start;
1122 			nblocks = seg_end - seg_start + 1;
1123 		}
1124 		kunmap_atomic(kaddr);
1125 		put_bh(su_bh);
1126 	}
1127 
1128 
1129 	if (nblocks) {
1130 		/* discard last extent */
1131 		if (start < start_block) {
1132 			nblocks -= start_block - start;
1133 			start = start_block;
1134 		}
1135 		if (start + nblocks > end_block + 1)
1136 			nblocks = end_block - start + 1;
1137 
1138 		if (nblocks >= minlen) {
1139 			ret = blkdev_issue_discard(nilfs->ns_bdev,
1140 					start * sects_per_block,
1141 					nblocks * sects_per_block,
1142 					GFP_NOFS, 0);
1143 			if (!ret)
1144 				ndiscarded += nblocks;
1145 		}
1146 	}
1147 
1148 out_sem:
1149 	up_read(&NILFS_MDT(sufile)->mi_sem);
1150 
1151 	range->len = ndiscarded << nilfs->ns_blocksize_bits;
1152 	return ret;
1153 }
1154 
1155 /**
1156  * nilfs_sufile_read - read or get sufile inode
1157  * @sb: super block instance
1158  * @susize: size of a segment usage entry
1159  * @raw_inode: on-disk sufile inode
1160  * @inodep: buffer to store the inode
1161  */
1162 int nilfs_sufile_read(struct super_block *sb, size_t susize,
1163 		      struct nilfs_inode *raw_inode, struct inode **inodep)
1164 {
1165 	struct inode *sufile;
1166 	struct nilfs_sufile_info *sui;
1167 	struct buffer_head *header_bh;
1168 	struct nilfs_sufile_header *header;
1169 	void *kaddr;
1170 	int err;
1171 
1172 	if (susize > sb->s_blocksize) {
1173 		printk(KERN_ERR
1174 		       "NILFS: too large segment usage size: %zu bytes.\n",
1175 		       susize);
1176 		return -EINVAL;
1177 	} else if (susize < NILFS_MIN_SEGMENT_USAGE_SIZE) {
1178 		printk(KERN_ERR
1179 		       "NILFS: too small segment usage size: %zu bytes.\n",
1180 		       susize);
1181 		return -EINVAL;
1182 	}
1183 
1184 	sufile = nilfs_iget_locked(sb, NULL, NILFS_SUFILE_INO);
1185 	if (unlikely(!sufile))
1186 		return -ENOMEM;
1187 	if (!(sufile->i_state & I_NEW))
1188 		goto out;
1189 
1190 	err = nilfs_mdt_init(sufile, NILFS_MDT_GFP, sizeof(*sui));
1191 	if (err)
1192 		goto failed;
1193 
1194 	nilfs_mdt_set_entry_size(sufile, susize,
1195 				 sizeof(struct nilfs_sufile_header));
1196 
1197 	err = nilfs_read_inode_common(sufile, raw_inode);
1198 	if (err)
1199 		goto failed;
1200 
1201 	err = nilfs_sufile_get_header_block(sufile, &header_bh);
1202 	if (err)
1203 		goto failed;
1204 
1205 	sui = NILFS_SUI(sufile);
1206 	kaddr = kmap_atomic(header_bh->b_page);
1207 	header = kaddr + bh_offset(header_bh);
1208 	sui->ncleansegs = le64_to_cpu(header->sh_ncleansegs);
1209 	kunmap_atomic(kaddr);
1210 	brelse(header_bh);
1211 
1212 	sui->allocmax = nilfs_sufile_get_nsegments(sufile) - 1;
1213 	sui->allocmin = 0;
1214 
1215 	unlock_new_inode(sufile);
1216  out:
1217 	*inodep = sufile;
1218 	return 0;
1219  failed:
1220 	iget_failed(sufile);
1221 	return err;
1222 }
1223